The invention discloses a multi-physical fused fire monitoring system and method thereof for predicting fire probability of occurrence. The monitoring system comprises a point type smoke sensing detector, a point type temperature sensing detector, a line type beam smoke sensing detector, an air-breathing smoke sensing fire detector, a cable line type temperature sensing detector, a cable multi-point type temperature detector, an in-place module, a first station layer switch and a cable fire monitoring early-warning system server. The method comprises the following steps: step one, orderly performing initialization, denoising and normalization on sensor data; step two, initializing a feature layer, selecting a random input weight beta and an output target T to form an input matrix H; step three, processing the feature layer data in batch, performing weight iteration after the initialization; step four, making the final judgment through an algorithm; and step five, judging that the fireoccurs and giving an alarm if a fusion result is greater than a set threshold, or repeating the steps 1-4. Through the system and method disclosed by the invention, the false alarm and the failed alarm of the fire can be effectively reduced, and the accuracy of the fire prediction can be greatly improved.