Framework of Electrical Fire Probability Estimation Based on Bayesian Network Model Inference

Abstract

Electrical fire had become one of the main parts in total fire accidents. Most of researches rely on the complex combustion models, which consume a huge number of computational resources. However, few studies focus on evaluating fire disaster by probability theory, and estimate the likelihood of fire occurring by the calculation result of probability based on the current data from the sensor. Bayesian Network is introduced due to the advantage of calculation complexity, ability of expressing uncertain factors and the accuracy of model with incomplete data. Some problems should be solved before using Bayesian Network to inference events based on given evidences. In this paper, the structure and the parameter of the Bayesian Network is created by the discussing result of the experts and scholars in electrical fire research field. A frequently-used fuzzy function called Sigmoid function to process data from raw data to the probability. Inference result by Bayesian Network is calculated by the Variable Elimination algorithm. A case study about the simulation of analyzing the probability of electrical fire happened when the load of circuit is under the high status. Research result shows that Bayesian Network model is suitable for estimating and analyzing in the scenario of electrical fire. Model has a good robust to express probability of electrical fire probability, which is of vital importance for estimating whether the fire occurs or not, thus providing significant information and instruction for preventing electrical fire and the sustainability of the environment. Based on the simulation result, it can conclude that the Bayesian network model inference is suitable for the electrical fire estimation scenario, and the introducing of this scheme is possible for predict electrical fire.