Study On The Influence Of Multi-Dimensional Characteristics Of Data On Fire Occurrence Probability And Trend Prediction

The prediction of fire occurrence probability and its development trend can contribute to allocate fire protection resources properly and take effective preventive measures to reduce the fire occurrence.Data processing is the key part for an accurate,efficient and universal fire prediction model.Therefore,this paper conducts theoretical analysis,model construction and application research on the influence of data types,quantity and interval on the probability and trend of fire,in order to establish an accurate,efficient and universal fire prediction model.The main research work and achievements are as follows:(1)Through numerical statistical analysis,the characteristics and main causes of fire in different geographical and time dimensions in China were obtained.Location: The fire situation in the eastern region is grim,and the fire is mainly inadvertently fired;The probability of fires in rural areas is large,and there are more fires caused by careless use of fire;Multi-storey building fires account for a relatively large proportion of them,which are easy to cause casualties.Time dimension: winter and spring are relatively frequent,electrical,arson and play fire fires are more concentrated;All kinds of fires are high at10～20 o’clock,and fires cause more casualties at 22～6 o’clock.From the above two aspects,the statistical analysis of 223 accident cases provides the basic data used to build the prediction model.(2)The optimal combination of data types for the fire predictive model is obtained through the control variable method,that is,the region,area and building type in the geographical location,the season in the time dimension,the electrical,arson,smoking,playing with fire,careless use of fire and production operations in the cause of fire.The prediction accuracy of the model using the best data is 92.86%,which is higher than 90.48%before screening.After sensitivity analysis,the regional,seasonal and electrical changes in the same data type had a great influence on the probability of fire,and the changes are0.90,0.47 and 0.88,respectively.(3)The influence of data volume and training ratio on the accuracy of the prediction model is studied.When the training ratio is 8:2,the accuracy of the prediction model gradually decreases with the decrease of the number of data.When the number of data remains unchanged,the accuracy increases firstly and then decreases as the training ratio increases.When the number of data is 223 and the training ratio is 8:2,the model prediction effect is the best,and the accuracy reaches 93.33%.After a sensitivity analysis,nine hazard factors requiring special attention,including regional,seasonal and arson,are identified.(4)The fire occurrence trend prediction model GM(1,1)based on different data intervals is established,and when the data volume Q = 3,the accuracy of the model built for quarterly data is the highest,99.98%;When Q = 8,the year data is the highest,at96.74%;When Q = 12,the monthly data is the highest,at 95.53%.The fire trends for the whole year,first quarter and first month of 2020 are 0.1660,0.9491 and 0.7335,which is consistent with the actual situation.