Forecasting System Based On Support Vector Machine Fire

Early prediction of fire to carry out a more complex and are of great significance to the research topic. Fire in the current prediction methods, based on traditional statistical forecasting methods fire prediction errors exist, and based on neural network prediction methods exist smart "over-fit" and the promotion of capacity is not strong, such as disadvantage. They are all false positives and omissions that may arise, such as serious consequences. To this end, this dissertation presents a support vector machine-based prediction model of the fire and complete the intelligent prediction system design and implementation. The main thesis job content and innovation are as follows:(1) Research support vector machines (SVM) algorithm principle,based on weighted support vector regression (WSVR) principle the fire detection methods. First of all,in describing the BP algorithm for neural network,LM algorithm neural network, RBF neural network based on that used in conventional fire detection deficiencies exist; And in discussing the basic principles of support vector machines and methods based on,In order to overcome the standard SVR does not take into account differences in the importance of the sample problem,using weighted treatment idea,put forward WSVR algorithm used for prediction of fire.(2) Design-based intelligent WSVR fire prediction system. Mainly include the sample selection and pretreatment information, prediction model training and parameter optimization of the design, prediction and output modules. Select the United Kingdom Institute of Boreham-wood fire experimental data simulation results show that the prediction based on the fire WSVR results are obvious,and is superior to the standard based on neural network and support vector regression prediction of the results, have good application prospects.The fire at the actual forecast, this dissertation presents the WSVR-based automatic fire detection technology, is feasible in the identification accuracy is better than on LM algorithm based on neural network methods and standards SVM algorithm has strong learning ability and generalization ability. Clearly, based on the prediction of fire WSVR intelligent system has broad application prospects.