Research On Dynamic Prediction Of Regional Landslide Risk Based On Cellular Automata

In recent years,the severe situation of landslide disasters has caused a huge impact on the social economy and the safety of the people’s lives,and regional landslide risk assessment is urgently needed.The landslide risk classification and zoning results obtained from the risk assessment can be used for the prevention and control of landslide disasters,and are of great significance for disaster prevention,mitigation and regional infrastructure planning.However,the traditional single landslide hazard assessment and deterministic physical model assessment methods are expensive and time-consuming to obtain data,and are not suitable for regional landslide assessment.Qualitative assessment methods such as analytic hierarchy process and expert scoring method require rich professional knowledge,and the assessment results are subjective.Quantitative evaluation methods such as generalized addition models,logistic regression,information methods,and support vector machines have the advantages of simple models and high efficiency,but they also have the disadvantages of simplifying complex geographic phenomena and less considering geographic spatial characteristics.According to statistical theory,it is usually required that the evaluation factors are strictly related or unrelated.In general,the previous evaluation methods have their own advantages and disadvantages,and there are few dynamic predictions of landslide risk.In view of the complexity,nonlinearity and uncertainty of geological landslide phenomena,from the perspective of research and application value,the dynamic prediction methods and models of landslide disasters are the goals that need to be studied and explored.Cellular automaton is an artificial intelligence method.As a discrete-time dynamic model of space-time,it provides a new way for the dynamic simulation and prediction of things or phenomena.However,due to the complexity of the geological landslide phenomenon,the direct application of the classical cellular automata technology to the dynamic prediction of landslide hazards has certain limitations.Convolutional neural network is a deep learning method that can automatically extract features based on input data,and has strong nonlinear learning and fitting capabilities.In recent years,it has been widely used in the field of pattern recognition and image classification.Case-based reasoning is a branch of artificial intelligence.It uses previous knowledge and experience to solve problems.For a complex new problem,you can get a solution to the problem without having to study the internal mechanism.Case-based reasoning is commonly used in environment science,planning,and other fields to solve problems such as classification,prediction,and diagnosis.However,the current method is mainly attribute-based reasoning,and less reasoning considering spatial features.This paper takes landslide as the research object,collected and analyzed the influencing factors and landslide data related to landslide development,and combined geographic information techniques and remote sensing techniques to analyze and excavate its spatial characteristics and distribution rules.Aiming at the problem of regional landslide risk assessment,considering the nonlinear relationship between landslide risk and various influencing factors,a convolutional neural network model with strong feature extraction ability was selected to evaluate the landslide risk in Lushan area from 2015 to 2018.On this basis,we focused on the study of geological cell space,cell state and dynamic evolution rules,established evolution rules based on spatial case-based reasoning and convolutional neural network,and finally constructed two regional landslide dynamic risk prediction models based on cellular automata.Using the spatial case-based reasoning based cellular automata model to predict the landslide risk in Lushan area,the convolutional neural network based cellular automata model as a contrast,the experiments obtained the regional prediction results of the dynamic prediction of landslide risk in Lushan area from 2016 to 2018.The accuracy of the results based on the convolutional neural network evolution rule from 2016 to2018 is 0.8908,0.8886 and 0.8873;the accuracy of the results based on the spatial casebased reasoning evolution rule is 0.8924,0.8933,0.8923.Finally,the spatial case-based reasoning based cellular automata model was used to predict the future landslide risk in Lushan area,and the landslide risk classification in 2020 and 2025 was obtained.Through experimental research and analysis,the paper draws the following conclusions.(1)Compared with the static landslide risk assessment method,the dynamic evolution prediction model based on cellular automata can achieve the equivalent effect,and the model can obtain the multi-year landslide risk assessment results through one data input.Therefore,for the dynamic prediction of the risk of regional landslides,the dynamic evolution prediction model based on cellular automata proposed in this paper is effective.(2)Aiming at the problem of dynamic prediction of regional landslide risk,the cellular automata model based on spatial case-based reasoning evolution rule is better than the model based on convolution neural network evolution rule to a certain extent.(3)In the distribution of landslide risk zones,the high-risk areas of landslides tend to shrink and tighten between 2016 and 2025,while the medium-and low-risk areas have slightly expanded.Because the high-risk areas are mainly distributed along the water system,road network and faults,the prevention and treatment of the Lushan landslide disasters in the later period should focus on the investigation and treatment of road slopes and river bank slopes.