冰雪天气下基于MFOA-KELM残差修正的跑道温度混合预测

道面温度短时精准预测是跑道积冰预警的关键因素之一, 为了解决单一机理预测模型随预测时间延长而造成误差累积的问题, 提出了一种冰雪天气下跑道温度混合预测方法。将跑道温度机理预测模型与核极限学习机(KELM)相结合, 建立一种数据驱动修正残差的跑道温度机理预测模型。针对果蝇优化算法(FOA)收敛速度慢、易陷入局部最小值的问题, 引入权值更新函数和距离扩充因子, 调整果蝇的全局寻优效果, 避免陷入局部极小值。利用改进的果蝇优化算法(MFOA)对KELM的正则化参数与核参数联合优化, 以冰雪天气下跑道温度实际数据为例, 建立基于改进果蝇优化核极限学习机(MFOA-KELM)的跑道温度混合预测模型, 并在不同时间尺度下对该混合预测模型进行仿真测试。实验结果表明:与单一机理预测模型相比, 当预测时长为120 min时, MFOA-KELM混合预测模型的平均绝对误差至少减小了61.43%, 在残差阈值为±0.5℃时, 平均预测准确率为91.25%。可见, MFOA-KELM混合预测模型具有更高的预测准确性, 研究结论显示该混合预测方法能够为机场跑道温度短时精准预测提供新思路。 

Runway temperature hybrid prediction based on MFOA-KELM residual correction under ice and snow

The runway surface temperature short-term accurate prediction is one of the key factors for runway icing warning.In order to solve the problem of error accumulation caused by a single mechanistic model with increasing prediction time, a hybrid runway temperature prediction method under ice and snow is proposed.The runway temperature mechanism model is combined with the kernel extreme learning machine (KELM) to develop a data-driven model for correcting the mechanism residuals. To address the problem that the fruit fly optimization algorithm (FOA) is slow to converge (converges slowly) and easily falls into local minima. By introducing a distance expansion factor and a weight update function, it is possible to modify the effect of the FOA's search for the global optimal solution and prevent falling into local minima. The modified fruit fly optimization algorithm (MFOA) is used to jointly optimize the KELM regularization parameter and the kernel parameter. A hybrid runway temperature prediction model is developed based on the modified fruit fly optimized kernel extreme learning machine (MFOA-KELM) with the actual data of runway temperature under ice and snow. The hybrid model is simulated and tested under different time lengths. The experimental results show that compared with the single mechanism prediction model, the mean absolute error of the MFOA-KELM hybrid model is reducedby at least 61.43% when the prediction length is 120 minutes, and the average prediction accuracy is 91.25% when the residual threshold is ±0.5℃. It can be seen that the MFOA-KELM hybrid model has higher prediction accuracy. The research findings show that this hybrid prediction method can provide a new idea for short time accurate prediction of airport runway temperatures.