The Distributed Ridgelet Kernel Model And Its Applications In Fire Scene Modeling

Fire signal is rather complicated one, which is a comprehensive reflection of combustion state and circumstances. The study of fire source single offers people an access to know the combustion rules, and also provides some effective approaches to prevent disaster and decrease loss. We can get multi-dimensional non-linear signal through the various types of detectors installed at different places of fire scene, which could reflect the real status of the combustion state. The advanced methods is used to research fire, which is able to make us to know and understand fire from different points, to dig out essential rules of combustion, to manage fire.By constructing non-linear system models, we forecast information of unknown points, for example, the temperature distribution of the fire scene, the intensity and location of the fire source etc., which is helpful to further understand the combustion state. There is a heated study on constructing complicated regression model of multi-dimensional non-linear system with machine learning method. In this paper, we propose an effective method of construct a multi-dimensional non-linear system—the distributed RKM. Starting from construction of kernel function and according to multi-resolution analysis, we analyze the feasibility and advantage of using ridgelet transform that has a good high-dimensional character as kernel function; Introducing the structural risk minimization principle from statistical learning theory to the method of training is to lessen the prediction error and improve the generalization; Combining the method of classification and cluster, the target vectors is divided into several parts, each which has its own corresponding models, in order to improve forecasting ability and robotization of theory. The method proposed in this paper is used to predicate the temperature in real fire scene, the intensity and the location of fire source and etc., which richens the information of the whole fire scene.Based on fuzzy adaptive resonance theory, multi-resolution analysis thought and structural risk minimization principles, we construct the model of distributed ridgelet kernel function regression, which is more suitable for multi-dimensional non-linear systems modeling. Both theory and practice prove that this model can accomplish multi-dimensional non-linear systems approximation and performs well. It is more appropriate for use in engineering context.