Research on Emergency Warning System for natural disasters Based on Multi-Sensor Information Fusion Jiang Shen, Tong-zheng Zeng, Tao Li, Hong Zhao* College of Management and Economics, Tianjin University, Tianjin, P. R. China (*zhaohong0129@163.com) Abstract - Accordance with these problems of the frequent natural disasters and the serious losses caused by natural disasters, the paper constructs a natural disaster emergency warning model and decision-making support system based on the research of multi-sensor technology and information fusion technology, and the functions of modules and system structures are also analyzed and studied. The research of this paper provides strong support on technique and system for the natural disaster decision-making emergency and warning management. Keywords - multi-sensor, information fusion, natural disasters, early warning I. INTRODUCTION In recent years, the frequently occurring of natural disasters highlights the environment vulnerability, resulting in a lot of property losses and casualties, the emergency management has been concerned by many researchers and policy makers. The devastating 9.0-magnitude earthquake in Japan in March 2011 leads to secondary disasters such as tsunamis, nuclear power plant leaks, causing serious casualties and economic losses; in 2010, Zhouqu, Gansu suffered a heavy rainfall which caused flash flooding and debris flow; in2008, China suffered snow disaster and Wenchuan earthquake which demonstrated the inadequacies of integrated emergency response system for natural disaster in China. Attentions should be given to the natural disaster detection and alarm systems after these events. With the application of multi-sensor information fusion technology of the emergency warning system for natural disasters, system information can be mastered more comprehensively and accurately, natural disasters can be monitored and pre-warned timely and accurately, the overall efficiency and effectiveness of existing emergency response system can be improved, and then the system is to bring greater social and economic benefits and better capabilities of the information sharing. This study can also promote the process of emergency management research in China, improve the level of natural disaster management and emergency response capabilities and provide a scientific basis and technical support of establishing the daily countermeasures and contingency plans for the authorities, making mitigation planning, deploying disaster prevention and relief work and making city development planning. II. RESEARCH STATUS AT HOME AND ABROAD AND TECHNOLOGICAL TRENDS A. Natural Disaster Research Natural disasters are the event that the variability of natural factors exceeds the bear or adaptability of human society, thus affecting human life, property and the living security. Essentially, natural disasters are the results that the earth's natural environment alters role in human society, including both the role of natural factors and the role of human society, especially the role of human society to withstand or adapt to the natural environment changes, namely, natural disasters have natural and social attributes[1]. Early researchers mainly associated natural disasters with hazard factors, they started natural disasters research with the growth characteristics and the formation mechanism of natural disasters such as typhoons, storm surges, heavy rain and floods and so on[2]. In accordance with the types of natural disasters mentioned by National Comprehensive Disaster Reduction "Eleventh Five-Year Plan", natural disasters in China can be divided into 13 kinds: floods, droughts, typhoons, hail, lightning, heat waves, dust storms, earthquakes, geological disasters, storm surges, red tides, forest and grassland fires, and plant diseases and insect pests. B. The application of information fusion technology Multi-source heterogeneous information’s multi-dimension, networking, dynamic characteristics requires information fusion technology to do multielement normalized fusion. Information fusion mainly includes two categories: fusion based on epistemology and fusion based on information theory, mathematical tools used in this field include probability theory, reasoning network, fuzzy theory and neural networks. In terms of seamless integration of multi-information and data association, Casasent[3], et al put forward Probabilistic Data Association (PDA), Bloon[4], et al proposed Neural Network Data Association (NNDA), Shalom[4], et al constructed interacting multi-model associated filter, Bozdag[5], et al put forward Joint Probabilistic Data Association (JPDA). NNDA, PDA and JPDA achieve the analysis of targets, sparse targets, high-density environment. In terms of multi-sensor information fusion theory, Waltz and Linas made comprehensive discussion on research, framework and application of the multi-sensor information fusion, Hall, et al made research on mathematical techniques in the multi-sensor data fusion, Bar-Shalom and Fortman[6] proposed new ideas and methods of multi-sensor data fusion in target tracking field. In the 1980s, Chinese scholars began to apply multi-sensor information fusion technology on national defense, military and other fields, after 1990s, multi-sensor information fusion technology began to be used in robotics and industrial process monitoring system. Information fusion is studied less on the control problems in the field of complex systems, in addition to the reason for the complexity of the multi-element and multi-dimensional information, the integrated issues on information fusion, learning mechanism as well as historical data mining are also very complex. III. INFORMATION FUSION TECHNIQUE Multi-sensor information fusion technology is a key technology of natural disasters monitoring and early warning in the decision support system, it joins all levels of various types of sensors located at each monitoring point into an effective network to achieve interoperability and comprehensive integration of information, of data acquisition and real-time monitoring tasks, the completion of each monitoring point. The role of multi-sensor information fusion system is the integration of multi-source information obtained by the various sensors to obtain more information than a single intelligence source and a better quality of information and information credibility, and greatly improve system access to information, the efficiency of the transfer, collection, processing, distribution and information applications [7]. The study selects applicable integration structure to complete the information fusion. Information fusion system architecture can be divided into three categories, namely, data level fusion, feature level fusion and decision level fusion [8]. Data level. It includes a lot of raw data collected by the multi-sensor system and the necessary signal pre-processing and analysis processes, such as signal noise removal, filtering, cross-correlation analysis, spectral analysis and so on. Feature level. It makes effective decision to data fusion results, it generally corresponds to the diagnostic methods of the natural disasters. Decision level. For the initial diagnosis of natural disasters, it supports decision-making to variety of countermeasures responding to disasters using decision fusion algorithms. On the functional characteristics, this three-layer structure can respectively meet the requests of monitoring alarm, disaster diagnosis and disaster countermeasures in the natural disaster decision support systems. Combining different sources of information or data collected by sensors in accordance with established rules can obtain more comprehensive and reliable awareness and understanding of the exact state of the object. The proposed rule methods include Classic reasoning method, Bayesian, Dempster-Shafer evidence theory, cluster analysis, expert system reasoning method and so on. IV. NATURAL DISASTER EMERGENCY WARNING SYSTEM MODEL BASED ON MULTI-SENSOR INFORMATION FUSION A. Information fusion model and method based on multi-sensor Because of information heterogeneity, uncertainty, multi-dimensional features of the complex system, we need to study the key integration and implementation issues such as multi-sensor technology, information fusion, adaptive control and so on; in decision-making and reasoning core technology, we need to study a variety of fusion strategies such as CBR (Case-Based Reasoning), RBR (Rule-Based Reasoning), realizing performance complement between CBR and RBR. The line of research preliminarily considered is realizing information collection, comparison and analysis by multi-sensor perception and integration, normalizing multi-parameter, and then on the basis of certain optimization criteria and the integration algorithm, realizing decision analysis and collaborative optimization using of CBR/ RBR integration decision reasoning model by the multi-layer fusion. B. Natural disaster emergency warning model based on multi-sensor information fusion To obtain information to support the final decision-making and control the disaster timely and accurately, natural disasters information is detected by multi-sensor in this model, then pretreated by filtering and normalization processing, and finally processed by multilayer fusion in the multi-layer information fusion center [9]. filtering processing Information Preprocessing human-machine interaction Information Fusion Center adaptive processing normalization of parameters data association Level 2: Information integration and optimization feature extraction Level 3: Information decision-making and alarming external data Information Fusion Decision Support System fault alarming danger alarming information sharing Natural Disaster Information Resources Database Level 1: information acquisition and preprocessing Multi-sensor Detection System manual intervention hazard information query disaster warning Fig.1 Natural disaster emergency warning model based on multi-sensor information fusion Level 1: information acquisition and preprocessing. The data, information and knowledge detected by multi-sensor form a multi-layer, multi-dimensional, heterogeneous and complex information space. Noise and interference signals in the complex information should be reduced before multi-sensors information processing. The pretreatment methods include filtering [10] and normalization. The multi-sensor information is filtrated by building a filter model, the uncertainty fire information data parameters and subspace features are normalized, so that multivariate information can be uniformly recognized and processed. Level 2: Information integration and optimization Though multiple natural disaster risk information has gone through the first level pretreatment, the essence of multiple heterogeneous information does not change. In order to achieve the fusion of these decision making knowledge which is uncertain, uncontrollable and heterogeneous in the data structure space, information fusion technology should be applied to build a multi-sensor information fusion model. Multi-sensor information fusion and optimization is done in the fusion processing center. In this center, multi-source information is processed in multiple levels according to certain rules and fusion algorithms. On the one hand, the information with the same purpose is connected [11], on the other hand, with the feature extraction of the associated data, a feature vector is obtained, and all feature vectors are merged so that we gain a target state or characteristics of target which are higher level than feature vectors. The results of the information fusion provide the basis and data support for natural disaster warning decision-making. Level 3: Information decision-making and alarming After multi-layer optimization of disaster information, the catastrophability still needs to be monitored by a decision support system to control disaster information, this level defines threshold of disaster information using multiple attribute decision making and multiple space scale analysis method, and then processes the information in the natural disaster warning system. This model puts the information processed as input factors of natural disaster warning decision system to definite threshold value, and gets system fault alarms, danger level forecast, and disaster alarming and response plan, etc. the system outputs information by multimedia hardware and software, and also provides inquiry through the man-machine interface. C. Decision-making information fusion support system based on Building information fusion decision support system via information fusion model is based on different information structure models, constraint rules, probability functions. The decision support system can be divided into four parts: information fusion algorithm (core model), human-machine interaction subsystems (components), database subsystem (data components), model library subsystem (model parts) [12]. (1) Information fusion algorithm Information fusion algorithm is multi-level progressive structure. It needs to select the appropriate information fusion algorithm model according to the information structure in the planning and design of the actual decision-making support system. Information fusion structure model generally includes centralized, hybrid, and feedback. (2) Human-machine interaction subsystem Human-machine interaction part is a bridge between natural disaster decision support system and system users, it is responsible for passing the decision-making information which is entered by users to the system for processing, and also passes the results processed by the system to users or requires users to enter information in accordance with system requirements. (3) Database Subsystem The database Subsystem mainly plays the role of storing data, maintaining data and ensuring data security and effective. The system stores data types include system user data, user decision-making demanding data, historical monitoring data, disaster parameter data, rule-based reasoning data and other data. (4)Model library and model library management system The model library section includes data pre-processing model, attributes reasoning model, disaster identification model, disaster early warning model, decision support program development model and so on. Model library Management System section is mainly responsible for creating, modifying and deleting the models. And also is responsible for defining the new model and making data exchange the databases. V. CONCLUSION Based on the analysis of the seriousness of natural disasters and the plague mechanism, the paper integrates hi-tech such as multi-sensor technology, information fusion technology, adaptive technology and so on, and builds decision-making model and its supporting system for natural disaster emergency warning which provide powerful technology and system support for natural disasters detecting and warning. The system guarantees to make more promptly and accurate decisions in the natural disaster warning system, and it has significant meanings for improving natural disaster emergency level and reducing disaster losses. ACKNOWLEDGMENT This paper is supported by National Natural Science Foundation of China (Grant No.71171143), Tianjin Research Program of Application Foundation and Advanced Technology (Grant No.10JCYBJC07300), Key Project of Science and Technology supporting program in Tianjin (Grant No.09ECKFGX00600), and FOXCONN Group REFERENCES [1] UNDP. Reducing disaster risk: a challenge for development. John S. Swift Co.,USA.www.undp.org/bcpr,2004 [2] Blaikie P., Cannon T., Davis I., et al..People's Vulnerability and Disasters [J]. London: Routledge. Natural Hazards,1994: 189-190. [3] Casasent, D. Yee, M. Optical data association neural net .Proceedings of the SPIE [J]. The International Society for Optical Engineering, 1993, 17(73):112-19. [4] Bloon H A P, Bar—Shalom Y. The interacting multiple model algorithm for systems with Markovi coefficients [J].IEEE, 2001, 33(8): 780-783. [5] Bozdag, C.E. Kahraman, C. Ruan, D. Fuzzy group decision making for selection among computer integrated manufacturing systems[J] Computers in Industry, 2003, 51(1):13-29 [6] Bar-Shalom Y, Formann T E. Tracking and data association [M].New York: Academic Press Inc,2003. [7] Yee-Ming Chen,Huang-Che Huang, Multisensor data fusion for manoeuvring target tracking[J].International Journal of Systems Science, 2001, 32(2):205-214. [8] S. Fabre, A. Appriou, X. Briottet. Presentation and description of two classification methods using data fusion based on sensor management [J]. Information fusion. 2001,(2):49-71. [9] Yong H, Jia Xing H, Ju Hua C. Software Project Risk Management Modeling with Neural Network and Support Vector Machine Approaches[C].Third International Conference on Natural Computation, Hainan, 2007:358-362. [10] Jose C. Geromel, Optimal Linear Filtering Under Parameter Uncertainty [J]. IEEE Trans. on Signal Processing, 1999, 47(1):168-175. [11] Jia, Y, Vithanage, C.M., Andrieu, C. Piechocki, R.J. Probabilistic data association for symbol detection in MIMO systems [J]. Electronics Letters, 2006, 42(1): 38-39. [12] Ford David T. Flood-warning decision-support system for Sacramento California [J]. Journal of water resources planning and management, 2001(4):254-260.