采用前向多层神经网络预测煤的自然发火期

煤自燃是煤氧复合的结果,在不同温度下煤氧复合的耗氧速率及CO、CO2产生率与煤的实验自然发火期之间存在复杂的对应关系,采用S型函数的前向多层人工神经网络来描述这种对应关系,用煤自然发火实验测定的数十个煤样的自然发火期及不同温度下耗氧速率及CO、CO2产生率作为训练样本,用BP算法对网络进行训练,得到了神经元间的联结强度.通过少量煤样程序升温氧化实验得到不同温度下煤样的耗氧速率及CO、CO2产生率,将其代入此人工神经网络程序就可以确定煤的实验自然发火期.该方法实验时间短、用煤量少得多,结果与实际吻合.

Prediction of Self-ignition Duration of Coal with Feedforward Multi-layer Artificial Neural Network

Self-ignition of coal results from coal and oxygen reaction. There is complex correspondence relation between experimental self-ignition duration and oxygen consumption rate, CO and CO2 generation rate of coal under different temperature. The correspondence relation was depicted by a feedforward multi-layer artificial neural network (ANN) with S-function. The ANN was trained by BP algorithm with tens of training swatches data from spontaneous combustion simulating experiments to determining self-ignition duration of coal, and oxygen consumption rate, the generation rate of CO and CO2 under different temperature, and connection strength of the neurons was obtained. Oxygen consumption rate, CO and CO2 generation rate of coal sample under different temperature were tested by programmed heating oxidation experiments with a modicum. And they were taken as input of the ANN program and self-ignition duration of coal can be calculated. It takes smaller amount of coal and fairly short period of time to obtain experimental self-ignition duration of coal with the given method, and the result accords with real condition.