The Non-stationary Time Series Prediction Based On Wavelet Analysis For Alumima Blending Process

The raw slurry blending is the first process of alumina production by sintering method. The quality of raw slurry will not only affect the alkali and water balance in the whole process, but also directly affect the quality of the final alumina product. So, it is vital to configure the raw slurry. In order to reduce energy consumption and improve the utilization rate of raw materials, the silicon slag is often returned to the blending by manufacturers in the process of the actual blending. However, the content of silicon slag with large fluctuation and the detection results with serious lag, make that the ratio can’t be timely and exactly adjustment. Hence, the qualified rate of the raw slurry is low. As a result, it is particularly important to focus on how to establish an effective prediction model.Based on a background of raw slurry blending process of alumina factory, this paper mainly focuses on the structure, analysis and forecasting of hybrid prediction model for the composition of silicon slag in the blending. The main research work is as follows:(1) According to the characteristics of time series of the returns silicon slag, three detection methods(i.e. the sample autocorrelation、DF and ADF) are used to judge the stationarity of the time series, then its short-term and long-term trends and patterns are analyzed. And the optimal wavelet function and reasonable wavelet decomposition level are selected so that the wavelet multi-scale decomposition is obtained.(2) According to the data feature of different scales, the different models are adopted for them respectively. For the high frequency data with great fluctuation, the classic ARMA model is adopted; For the high frequency data with gentle fluctuation, BP neural network model is used; And the non-seasonal Holt-Winter model is built for the low frequency with obvious. These model parameters are optimized and the corresponding sequences are forecasted.(3) The hybrid prediction model is proposed by reconstructing all forecast data in different frequency. And the prediction results are compared with the results of other prediction methods.Forecast and comparison results show that the hybrid prediction model based on wavelet analysis can quickly forecast the composition of silicon slag with better prediction accuracy. Therefore, it is a great practical potentiality in alumina production process.