Mine underground adaptive fire grading early-warning method and system

Abstract

The invention provides a mine underground adaptive fire grading early-warning method and system. The mine underground adaptive fire grading early-warning method comprises the following steps that S1,marked sample data are selected for BP neural network training; S2, a BP neural network model is optimized through a migration learning strategy, and an output layer parameter subjected to fine-adjusting training is obtained; and S3, the collected parameter value is input, and the fire early-warning grade corresponding to the parameter value is output through the optimized BP neural network model.In the step S2, the BP neural network model is optimized through the migration learning strategy, then the judgment capability during a fire can be preserved, the non-fire situation in a mine can also be distinguished, thus the accuracy rate of fire grading early-warning of the mine is increased, the situations of false report and report missing are reduced, and the adaptive function of fire grading early-warning in complex environments such as the mine and a construction tunnel is effectively achieved.

Description

technical field [0001] The present invention relates to a fire early warning method, in particular to an underground adaptive fire grading early warning method, and a fire grading early warning system adopting the underground adaptive fire grading early warning method. Background technique [0002] Using BP neural network for fire early warning, firstly, it is necessary to carry out parameter training on the built BP neural network model, so that the model can finally output the probability of the current fire occurrence according to the three data input by the sensor. In the field of fire early warning, there are two public data sets of Chinese standard test fire and European standard test fire. We usually use this kind of data as the training set to train the network model. [0003] However, such data is usually measured in a laboratory or in a specific environment, and the network model trained by it is often not suitable for complex environments. For example, under the ...

AI Technical Summary

Problems solved by technology

Therefore, if the model is fitted only by adjusting the threshold, the accuracy is often not accurate enough

If we want to get a network model suitable for downhole, we need to obtain parameters such as temperature, smoke concentration and CO content when the downhole simulated fire occurs to train the model. Can't achieve

Images