Research On Fire Warning System Based On Multi-source Information Fusion

With the continuous improvement of people's living standards,urban buildings and population density continuously increase.Various electrical and wire networks are intertwined,which increases the hidden fire risk.It is of great significance to the protection of personal and property that perceive the signal changes and make correct countermeasures before the fire.Traditional fire warning methods include threshold method,trend method,etc.These methods usually consider the change of one characteristic.For complex and changeable fire environments,there are often false alarms and missed alarms.With the popularization of artificial intelligence technology,intelligent algorithms,such as neural network algorithm,are gradually applied to fire prediction.These algorithms have good fire recognition performance.However,there are still limitations in terms of algorithm understanding,hardware implementation,cost control,etc.Aiming at solving these problems,a novel fire warning method based on multi-source information fusion is developed in this paper.A Double Weighted and Coefficient Compensation Naive Bayes(DWCNB)method for early fire warning is proposed.The relationship between the characteristic attributes is considered to double weight the characteristic attribute and the value of the characteristic attribute.The assumption that Naive Bayes' attribute independence and have same importance is weakened.At the same time,the method of compensating the prior probability is proposed to balance the decision degree of the prior probability and conditional probability.The main research contents include:(1)The Naive Bayesian algorithm is analyzed and improved,including Laplacian smoothing,logarithm operation.By considering the relationship between characteristic attributes and the attribute values,the double weighting method are conducted,and Naive Bayes' conditional independence assumption is weakened.The compensation coefficient is used to compensate the prior probability,and the proportion of decision-making is balanced.(2)The Python language is used to build the improved algorithm model,and its performance is evaluated and analyzed.The compensation coefficient is designed by the orthogonal test.The proposed method is compared with several commonly used machine learning algorithms.(3)The improved classification model is implanted into the embedded device,and the STM32L151 chip is used for control.The temperature,smoke concentration and carbon monoxide concentration signals are sampled by the DHT11 temperature sensor,the MQ-2 smoke sensor and the MQ-7 carbon monoxide sensor respectively.The NB-Io T module BC28 is used for wireless communication.The software includes improved Naive Bayesian algorithm,ADC analog-to-digital conversion,USART communication,and NB-Io T communication.(4)The experiment of fire warning is designed.Four kinds of test fires and three kinds of interference sources are tested for the fire early warning.And the results are compared with the market general smoke alarm.The simulation and experimental results show that the accuracy of the proposed algorithm is up to 98.13%.The average accuracies of test fire and interference source are 97.76% and 98.24%,respectively.