A Fire Hazard Ranking Distribution Determine System Based On Multisensor Information Fusion

Fire accidentsareamong the major disasters in human life and have caused catastrophic results in many situations. There has always been a battle between fire and human beings. The development of electronic engineering, communication technology and computer science has given humans powerful tools to help achieve success in the battle. With pre-allocated distributed fire monitoring systems in important buildings, detailed and precise information can be provided for fire emergency management and firefighting after a fire accident. However, the current prevalent fire detection system can merely give a fire alarm and will be ignored in the ensuing fire emergency management process. On the one hand, high false alarm and failed alarm ratesoccur, due to the application of a single fire sensor type, which merely offers a partial description of a fire. Therefore, multi-sensor technology is needed to provide accurate fire detection results. On the other hand, though using multiple sensors to measure all parameterscan depict a fire event in a building in detail,a large flow of data are difficult to interpret and can rapidly result in information overload. Thus, some intelligent fire information management system employing multi-sensor technology in addition to the current fire detection systemshould be developed.The primary objective of this research is to establish a multi-sensor fire hazard ranking distribution determine system that can assist fire monitoring, fire emergency management, fire rescue and firefighting. With fire information from a variety of sensors, the system can provide the necessary fire hazard ranking for distributed areas in a building and redundant information can be excluded with information fusion to supply thenecessary fire information for fire emergency management, fire rescue and firefighting.Firstly, a multi-sensor fire signature combination selection model is proposed to determine effective fire signature combinations in multi-sensor fire detection. This approach uses mutual information entropy theory with the concept of maximum relevance and minimum redundancy. In contrast to traditional experimental measures to conduct a large number of fire experiments with different fire signature combinations, this model can obtain the necessary results with limited experiment and thus reduce time and cost requirements. Secondly, a Fuzzy Full Raw Data (FFRD) feature extraction algorithm is proposed to generate inputs for classifiers in multi-sensor fire detection. This algorithm has the potential to generate supervised training data for artificial neural network modelsfrom limited experimental results. A dynamic observation window is used to extract the necessary multi-sensor fire information. Sensitivity analysis of step size and window duration of the feature extraction window is investigated. Multi-sensor fire detection performances of several reviewed well-known Artificial Neural Networks in multi-sensor fire detection, including Back Propagation, Radius Based Function, Learn Vector Machine and Probabilistic Neural Network, are investigated.Thirdly, a multi-sensor fire detection model based on the FFRD algorithm and the investigation results outlined above is proposed. The model comprises three modules,namely a fire signature combination selection module, a supervised training module and a fire detection module. Some large scale experiments were conducted in an ISO9705fire test room with different polymer materials toverifymodel validity. Diverse fire detection results confirm that the proposed multi-sensor fire detection model offers satisfactory performance in terms of fire detection sensitivity and reliability. It is demonstrated that the model has the capacity to limit the influence of fluctuations in measured fire parameters and has satisfactory fault tolerance capacity.Fourthly, a conceptual fire node network algorithm is proposed to transfer a building plan into a fire node network. Fire hazard ranking of each fire node can represent the results of a control unit related to a protection area in the building plan. From the multi-sensor fire detection results, the fire hazard ranking of each fire node can be determined in relation to three correction coefficients, namely fire status, fire source distance and multiple detected fire spots.Finally, a fire hazard ranking distribution determine system including a fire node subdivision module, a multi-sensor fire detection module, a fire information cloud, a fire hazard ranking module and a fire hazard ranking distribution merge module is proposed for fire emergency management. Several remote transfer network topologies within the application of the system are presented. Then, the potential long term significance of the proposed fire hazard ranking distribution determine system is discussed.