Research On Displacement Prediction And Risk Assessment Of Landslide Based On Intelligent Algorithm

Landslide hazard is a common geological disaster and its occurrence not only brings serious threats to human life and property safety but also causes huge damage to resources, environment and ecology and other aspects. China is one of the countries suffering the most serious landslide hazard in the world. According to China geological disaster report issued by China Geological Environment Information Network, there were 260,353 geological disasters during 2006 and 2015 in China and the landslide occurring ratio occupies the first place of geological disasters, accounting for 73.79%. During these 10 years, landslide led to casualties and missing people of 11,281 and direct economic loss of 43.505 billion Yuan. Moreover, secondary disasters caused by landslide also are hard to estimate. Therefore, it is necessary to take measures to monitor them and forecast landslide hazard scientifically and effectively, which is of great economic value and social significance.However, there are numerous influence factors of landslide development(for example, topography, geological structure, formation lithology, hydrogeological conditions, rainfall and human engineering activities), which lead to openness, complexity and uncertainty of landslide movement. These make it difficult to accurately forecast for traditional methods. Therefore, this paper proposes intelligence algorithm as the main research means and integrates modern surveying and mapping data processing, grey prediction, grey decision and other theoretical knowledge to study typical landslide projects in China, Lianziya landslide, Wolongsi landslide, Gushuwu landslide, Xintan landslide, and Chongqing Wanzhou District landslide. With topics of landslide deformation displacement forecast and landslide hazard class division, intelligence algorithm-based landslide disaster forecast system is constructed and its major study contents and achievements are as follows:(1) Landslide deformation displacement condition recognition and curve characteristics classification(1) For landslide mass in the nature, its formation conditions and inducing factors differ, which lead to different patterns of landslide deformation accumulative displacement curve. Characteristics of the landslide deformation accumulative displacement curve are studied thoroughly and according to patterns, it is divided into four types, decelerating- uniform velocity, uniform velocity-accelerating, deceleratinguniform velocity- accelerating and compound type. Based on the analysis of topography, geological structure and influence factors of four typical landslide projects, Lianziya landslide, Wolongsi landslide, Gushuwu landslide and Xintan landslide, characteristics of their deformation curves are recognized and classified.(2) Mastering characteristics of the landslide deformation accumulative displacement curve is of great guidance for qualitatively determining the genetic model of landslide, deformation development stage, influence degree of inducing factors and selection of prediction and forecast model.(2) Classic intelligent algorithm BP and RBF-based landslide deformation displacement prediction(1) Considering slow network convergence speed of standard BP algorithm and being easy to fall into local minimum, four types of BP improved algorithms are derived. Taking Lianziya landslide, Wolongsi landslide, Gushuwu landslide and Xintan landslide as an example, BP improved algorithm-based landslide deformation displacement prediction model is established and several problems of modeling BP algorithm are discussed thoroughly. Also, the detailed implementation process of BP network structure parameter optimization is given and the optimal network topology of BP landslide deformation displacement prediction is constructed. The example indicates that these four improved algorithms have obvious improvements in forecast effects compared with standard BP algorithm and LM-BP algorithm owns the optimal forecast effects.(2) RBF neural network structure, training algorithm and modeling progress are illustrated and 2D range query-based RBF network parameter optimization method is proposed. RBF and LM-BP are compared in hidden-layer transfer function, hidden-layer node number, training algorithm and approach method. Taking Lianziya landslide, Wolongsi landslide, Gushuwu landslide and Xintan landslide as an example, the applicability of RBF and LM-BP algorithms to landslide deformation displacement prediction is analyzed and compared. The experimental results show that compared BP algorithm, RBF has improvements in network convergence speed, network generalization ability and extrapolation forecast.(3) New intelligent algorithm ELM-based landslide deformation displacement prediction(1) ELM algorithm is introduced in landslide deformation prediction to thoroughly analyze the learning mechanism of ELM, pointing out the its essential difference from BP and RBF algorithm. ELM solves network local minimum of classic BP and RBF algorithms, which adopts gradient descent training network.(2) In view of error processing, the solution of ELM network output weight parameter β? is derived and it is found that its solution procedure is based on least square estimation, leading to poor resistance to gross error of ELM algorithm.To enhance the resistance to gross error of ELM algorithm, generalized maximum likelihood estimation(M estimation) is integrated with it and M estimation-based Robust-ELM landslide deformation displacement prediction is proposed.(3) Standard ELM algorithm is sensitive to gross error in landslide data and owns poor resistance to gross error. M estimation-based Robust-ELM algorithm can better resist single and multiple gross errors in landslide data and its prediction accuracy is high.(4) Intelligent coupling model based landslide deformation displacement predictionFor problems of single intelligence algorithm-based landslide deformation prediction, landslide coupling prediction model is raised. Based on three views, three-form coupling models are constructed, weight restriction-based coupling prediction model, algorithm fusion-based coupling model, and coupling model considering the influence of inducing factors.(1) According to different constraint criteria of weight, this paper studies the construction of weight coupling prediction model, respectively constructing five-form weight coupling prediction models, optional weight, non-optimal weight, grey comprehensive correlation, fix weight, entropy weight and ELM non-linear weight. The solving characteristics of these five constraint criteria weight are discussed. Taking Gushuwu landslide and Xintan landslide as an example, the coupling forecast effects of these five constraint criteria are compared and the examples demonstrate that ELM-based non-linear coupling prediction owns good characteristics and high prediction accuracy.(2) When algorithm fusion-based coupling prediction model is constructed, first of all, gray accumulation generates operators and weakens randomness of landslide deformation displacement data. Podzolic horizon-based ELM coupling prediction model is constructed. Second, considering the effective information provided by grey model group prediction, grey model group-based coupling prediction model is constructed.(3) Taking Xintan landslide and Three Gorges Reservoir Region landslide as an example, multi-factor coupling prediction model considering the influences of rainfall, reservoir water level and inducing factors is established. This model firstly decomposes landslide deformation displacement into trend term and stochastic term and utilizes GM(l,l) model to extract trend term of deformation. Then, stochastic term of ELM neural network forecast displacement based on rainfall, reservoir water level and inducing factors is established. The example indicates that this coupling model can combine the characteristics of landslide monitoring data and integrates data trend and randomness starting from data decomposition. At the same time, the inducing factor of landslide mass is considered. It fully utilizes the effective information of landslide monitoring data and owns high prediction accuracy.(5) Multi-factor weighed grey target decision theory-based ELM landslide danger evaluationTaking Chongqing Wanzhou district landslide as an example, the basic evaluation idea of landslide formation mechanism analysis- major evaluation index selectionevaluation index quantization- mathematical model construction- landslide hazard class division is established. Firstly, multi-factor weighed grey target decision theory is introduced in the landslide hazard assessment and it is coupled with a new intelligent ELM algorithm. In this way, multi-factor weighed grey target ELM landslide hazard assessment model is constructed.(1) Starting from landslide hazard formation conditions and inducing factors, elevation, slope, sliding mass material types, rainfall, human engineering activities and other 11 factors are selected as landslide hazard evaluation indexes. Grey correlation analysis is utilized to obtain the weight of influence factors.(2) On the basis of grey target decision theory, quantitative evaluation is made on landslide hazard level according to off-target distance. The landslide hazard level is classified into high, relatively high, middle, low and relatively low. Furthermore, ELM algorithm is adopted to forecast the off-target distance of to-be-evaluated landslide mass, realizing the hazard class division according to off-target distance.(3) When landslide hazard class division is made by weighted grey target decision-based ELM model, it comprehensively considers the effective information provided by different factors and the landslide influence factors and weight allocation. The quantization of hazard level is realized by comparing off-target distance. According to the comparison between evaluation data off-target distance and standard target center, landslide hazard level is classified. Thus, multi-factor qualitative and quantitative prediction is achieved, which improves the scientificity and accuracy of landslide hazard prediction.