中文摘要：

分析了卡尔曼滤波预测空气质量指数的机理，用ARMA算法为卡尔曼滤波建立模型，提出了将RBF神经网络融合于卡尔曼滤波的方法，实现对空气质量指数的混合预测。根据空气质量指数时间序列的特点，建立了自回归预测模型，进而建立卡尔曼滤波的状态方程和测量方程。采用随机梯度逼近训练算法训练RBF神经网络，用RBF神经网络的输出作为卡尔曼滤波测量方程的观测值。融合了RBF神经网络后的卡尔曼滤波预测方法减少了单一方法的延迟现象，使同种性质的误差累积减小，提高了预测精度。对AQI序列预测的仿真显示融合后的卡尔曼滤波方法优于单一的卡尔曼滤波方法，亦优于现已广泛应用的BP神经网络预测方法。

英文摘要：

The prediction mechanism of Kalman filtering for air quality index was analyzed. And a hysteretic neuralnetwork was proposed to predict the air quality index series. State equation for Kalman filter was established by AR-MA model. The hybrid prediction of air quality index, combining Kalman filter and RBF neural network was pro-posed. According to the characteristic of air quality index series, autoregressive model was established. And then,the measurement equation and the state equation of Kalman filter were established as well. Stochastic gradient ap-proximation method was applied to train RBF neural network. The output of RBF neural network was regarded theobserved value by Kalman filter. Hybrid prediction＇ s main advantages included preventing forecast delay causedby the single prediction mechanism and precise forecasting. The results of predicting air quality index simulationshowed that the hysteretic Kalman filter had better prediction performance than original Kalman filter, and the hys-teretic Kalman filter was also superior to BP neural network.