Forecasting Of Debris Flow In Panshi City Of Jilin Province

China is one of the most serious nations threaten by debris flows in the world. The debris flow disaster results in the death of hundreds of people average annually and causes a direct economy loss of more than 1 billion RMB, seriously threaten national economy and social continuable development. At the present time, debris flow cannot be harnessed all-around because of its genetic complications, great capacity and widely distribution and cost much on harnessing. As an important disaster decreased measure, forecasting of debris flow comes under abroad attention of scholars inside and outside nation, and becomes hot spot and sixty-four-dollar question on the disaster decreased research.Debris flow occurrence scale not only is an all-important parameter that can be directly used for preventing engineering design, but also related ultimately settlement about the forecasting of debris flow dimensional small scale, at the same time it intimately mutuality with debris flow occurrence frequency and dangerous degree assessment. Debris flow occurrence scale forecast is the emphasis on the research of debris flow forecast, and it is also a cosmopolitan difficult problem that all countries scholars applied themselves to take. Essentially, Debris flow occurrence scale forecast is a kind of identification model. Therefore, using artificial neural network (ANN) model to forecast debris flow scale comes into existence in logic.The artificial neural network has extensive application in many fields. It has much application in the field of engineering geology too. But the method is far immature and existing defect not only in theory but also in practical application. So it is very significance to study the method and to create a more rational and better utility artificial neural network.The thesis bases on the work of geological hazard survey item"The Survey and Zoning of Geological Hazard in Panshi City of Jilin Province" issued by ministry of Land and Resources P.R.C, using ANN Back-Propagation (BP) artificial neural network into the research of debris flow forecast in order to find a reasonable debris flow scale forecast model, and process a debris flow scale forecast in investigation area to satisfy the demand of practical work.The thesis introduced physical geography general situation of Panshi City, and based on the analytical mechanism of debris flow disaster's cause of formation in investigation area, batted around the relations between each kinds of genes and debris flow, such as weather and hydrology, landform and physiognomy, stratum and lithology, geology and conformation, hydrological geology, earthquake activity, mankind activity and so on.The thesis exerted itself to discuss the ANN applying in the forecasting of debris flows.The thesis chose 8 main factors that influence debris flows developing as influencing factor, and carried on grades to debris flows' scale according to the investigation area's actual conditions. The result of grading was propitious to train the BP neural network. There were 166 representative data swatch selected from 199 observation data of Panshi City debris flows, which used for establishing the artificial neural network model, another 9 were chose as test swatch, and others were forecast swatch.The thesis studied the selection of the number of implication layer neuron. The results indicated that when the multiple of the number of implication and importation layer neuron was 1-2, were all suitable to the net, then the net capability became better if the number of implication layer between 15 and 19.the thesis selected 17 as the number of implication layer to optimize the net. The thesis chose 13 different algorithms to train the net. During the net training processing, most algorithms would meet the local minimization problem. If adopt the optimize ways to improve. Commonly, this problem can be avoided from high frequency by adopted the optimize way to improve the net, such as random initializing quantity tentatively quantity and threshold quantity, using the train algorithms with momentum, amending important training parameter and so on. According to compare with 13 functions, found 8 suitable arithmetic functions.The thesis selected preference 3 algorithms from the 8, and corresponded to establish 3 kinds of forecast models that were model (1), model (2) and model (3). Model (1) named Levenberg-Marquardt BP net algorithm well and truly forecasted 22 data, and its emulator correct ratio was highest, to 73%, and its mean-squared error was 0.1276.The thesis made use of model (1) to set up debris flow forecast model, which was with rapidness training speed, highness precision, smallness emulator error, and forecast values were close to measured value. All accounted for that BP net is appropriate to be applied in geological hazards forecast, and has wider applying foreground.To sum up, applying BP net to forecast debris flow scale can put up strongly system stability, recipient and anti-interference, and its forecast process is simple and speed.