Analysis On The Characteristics Of Soil Water Content Time Series Based On Fractal Theory

Researching on the internal change law of soil water content time series in time and space plays a very important role to further predict soil water content. O n the basis of fractal theory, the time series of soil water content of three representative levels in Luancheng district are selected as the research object; the change characteristics and fractal characteristics are analyzed and studied. And a regressive model based on chaotic least squares support vector machine is built to predict soil water content. This thesis has achieved several results as following.(1) Kolmogorov-Smirnor normality test is introduced to calculate the statistical characteristic values of time series of soil water content. The results indicate that the overall distribution of different levels time series does not comply with the characteristic of random normal distribution, the characteristic of “peak and fat tail” of time series are proved. So the time series can be judged to have the feature of nonlinear and fractal distribution.(2) The Hurst index and fractal dimension are respectively calculated to study its long-term sustainability features and the amplitude of the fluctuation by using R/S. It can be seen from the results, the Hurst index of three levels are greater than 0.5, which reveals that the different levels of soil water content time series is nonrandom and persistent and there is significant long-range in accordance with the time series. The fractal dimension are between 1 and 2, and they become smaller and smaller with the increase of the soil depth. Results show that the fluctuation rate of soil water content tends to small and less noise with the increase of soil depth.(3) By applying the detrended fluctuation analysis, from the perspective of scale invariance, the long-range dependence of the time series of the 14 levels of soil water content is analyzed, and the fractal characteristics of soil water content in space is revealed by fractal dimension. The result shows that the 14 levels soil water content time series have long-range correlation and obvious trend change. The scaling exponents is increased gradually with the increase of depth, it reveals that the long-range correlation of soil water content change is more and more strong; O n the other hand, fractal dimension is gradually become smaller, it reveals that the volatility of soil water content is more and more stable, which accords with the actual situation.(4) This thesis focuses on the 10 cm soil water content time series, the best time delay and embedding dimension of time series is respectively calculated by using the self correlation coefficient method and the G-P algorithm, which are used to reconstruct the phase space. And judging the time series has chaotic characteristics by computing the largest Lyapunov index and the correlation dimension. Based on this, the predictive model of chaos- least squares support vector machine is constructed to further predict and analysis soil water content. And according to the evaluation index parameters of the prediction model, the accuracy and reliability is tested. The results show that the predictive model is effective and believable.