The invention provides a hot-rolling coiling temperature forecasting method based on a relevant neural network. Data of strip steel thickness H, rolling velocity V, finish rolling temperature T1, strip steel width W, super rapid cooling valve opening quantity N1, laminar flow valve opening quantity N2, super rapid cooling inlet temperature T2 and target coiling temperature T3 are collected through a personal digital assistant (PDA) terminal, an input matrix [H, V, T1, W, N1, N2, T2, T3] is built, and an output matrix [T] is built. If [V, T1, W, N1, N2, T2, T3] is supposed to be not changed, Y is defined to be an influence degree of the H. According to influence degrees, corresponding influence degrees are given to the weight of the input end and a hidden layer and to the weight of the hidden layer and an output layer, three layers of ASBP neutral networks are built, the actual coiling temperature is output, ASBP neutral network training is conducted, and coiling temperature forecasting is conducted by using actual testing data. By means of the method, a coiling temperature forecasting error range is reduced from -20 DEG C-20 DEG C to -10 DEG C-10 DEG C, and coiling temperature control is more accurate.