改进PSO-GA-BP的PM2.5浓度预测

针对空气中PM2.5浓度受到气象因素和大气污染物的影响,且具有非线性、不确定性等特征,提出BP神经网络的预测方法。利用粒子群优化思想,在寻优过程引入遗传算法的交叉和变异操作,设计一种改进的PSO-GA混合算法对BP初始权值和阈值进行设定,有效避免陷入局部极小,提高收敛速度。仿真结果表明,改进的PSO-GA-BP预测模型和PSO-BP预测模型均可获得良好的预测结果,它们的RMSE、MAE和MAPE相差不大,分别为8.961、6.974、0.140和9.561、7.226、0.146,但在设定相同的进化代数时,改进的PSO-GA-BP预测模型比PSO-BP预测模型收敛性更好。

PM2.5 concentration prediction based on improved PSO-GA-BP

The PM2.5 concentration is influenced by meteorological factors and air pollutants with the characteristics of nonli- nearity and uncertainty. The prediction method of BP neural network was proposed. To solve the problems of slow convergence and falling into local minimum, a hybrid algorithm based on genetic algorithm (GA) and particle swarm optimization (PSO) was designed to define the initial weights and thresholds of BP neural network. The crossover and mutation in GA were introduced to the PSO. The simulations demonstrate that the improved PSO-GA-BP model and PSO-BP model both can obtain good prediction results, their RMSE, MAE and MAPE are 8.961, 6.974, 0.140 and 9.561, 7.226, 0.146 respectively. But when the evolutionary generations is the same, the convergence of PSO-GA-BP model is better than that of PSO-BP model.