Study Of Groundwater Depth Space-time Forecast Methods Based On Hybrid Model

Arid area Minqin oasis natural conditions are poor,water resources are poor,weak ecological environment.Recently,the function of groundwater level has been seriously degraded,ecological problems are frequent,Minqin oasis has become a typical research area.how to make scientific,accurate and timely grasp of arid and water resources in semi-arid areas become a key problem in water resources research in such areas.The area of groundwater depth is the time and space variable that changes with the change of location and time,the purpose of spatial and temporal prediction of regional groundwater depth is to reveal the dynamic changes of groundwater distribution and distribution over time in the future,and it is the artificial dynamic simulation of groundwater depth.Spatio-temporal data analysis provides important theoretical support for the construction of regional groundwater depth-time prediction agent model.Spatio-temporal prediction is the basic problem of spatial and temporal data analysis of regional variables.In recent years,it has become a hot research topic both at home and abroad.Spatio-temporal prediction is based on the time-space sequence prediction theory to establish the mathematical model.Based on the time-domain prediction,the future prediction surface is established,and the future distribution of groundwater depth and its temporal and spatial variation are revealed.The future distribution of groundwater depth is predicted.The temporal and spatial prediction of groundwater depth is different from the traditional time series prediction and simple spatial prediction.How to improve accuracy is the core problem of forecasting.At present,the commonly used models have insufficient consideration for the complex time and space effects of different monitoring sequences,and the lack of nonlinear consideration is not enough to influence the accuracy of the model prediction caused by different parameter determination methods.Hybrid prediction is a combination of different models,different time series analysis theory,fusion or integration,constitute a prediction model,is to combine different models together to predict,only one prediction results.The hybrid prediction model is an important and frontier research method to improve the prediction accuracy.At present,the hybrid forecasting model based on nonlinear theory,multivariate time series analysis theory and parameter heuristic optimization algorithm is the main technical means to improve the prediction accuracy.In recent decades,with the oasis of agricultural development of a large number of groundwater resources development and utilization,so that the function of the groundwater system serious decline,oasis shrinking,triggering a series of ecological and environmental problems.Minqin groundwater research has caused widespread concern and attention from all circles at home and abroad.This paper uses the correct and scientific method to study the groundwater resources of Minqin Oasis,complete the simulation of groundwater depth in Minqin County,establish the regional groundwater spatial forecasting system,and make important theoretical and practical value for regional groundwater research.Problems encountered in the study:(1)missing data repair problem;In the long-term monitoring process,due to human reasons and hydrogeology and other reasons,resulting in two types of missing data: the same site,intermittent detection caused by intermittent missing data;groundwater monitoring wells abandoned,transposition made of new truncated missing data.In traditional research,often spliced sequence directly before and after the transposition.Due to differences in hydrogeological conditions and differences in soil vegetation,the differences in detection values before and after transposition of the monitoring wells are neglected.This difference may change the inherent continuity and temporal and spatial correlation of the original sequence,resulting in a larger prediction result error.It is a challenging task to study the establishment of a large sequence of truncated long sequence missing sequences.(2)Lack of efficient,high-precision time-space multi-sequence prediction model problem;The time domain prediction of regional groundwater depth is achieved through the prediction of time and space sequences obtained by long monitoring of multiple monitoring points.The key to improve the accuracy of time and space prediction is to improve the accuracy of prediction and spatial interpolation of the monitoring sequence.The traditional prediction model is mainly based on time series analysis theory,through has established temporal-spatio sequence analysis theory,neural network and machine learning method.These methods have great workload,and its have poor adaptability to multi-monitoring points and large regions,and the prediction accuracy is not high.(3)It is poor that the accuracy and commonness of Spatial interpolation method;The ultimate goal of temporal-spatio prediction is to visualize the future pattern of groundwater depth.Past spatial interpolation methods such as: surface fitting method;spline interpolation method;Kriging interpolation method successfully depicts the spatial pattern of groundwater level,but these methods only consider the known monitoring point of the measured value of the estimated value Space influence,without considering the impact of historical data on the valuation,temporal-spatio Kriging interpolation takes into account the temporal and spatial correlation,but its key temporal-spatio covariance function is established,the versatility is poor and the precision is not high.In this paper,the nonlinear analysis method of the least squares support vector machine(LSSVM),generalized regression neural network(GRNN)and the new artificial intelligence parameter optimization method,grid search(GA),Cross-validation(CV),self-organizing neural network mapping(SOM)clustering,wavelet denoising,This paper has carried on the thorough research to the temporal and spatial prediction of groundwater depth in dry Minqin oasis,and has carried on the following research work:(1)Study on mixed model of space-time restoration of missing data;In order to obtain a more scientific and reliable groundwater depth study of temporal and spatial sequences,this paper proposes a mixed model of SOM-FLSSVM space-time repair,which is worthy of repair in space-time sequence.The model not only considers the time factor of each type of missing data monitoring site,but also takes into account the spatial factors,make full use of missing data to monitor the site and its adjacent site information for data interpolation.Experiments show that the proposed model of space-time prediction compared to some other classic data interpolation model,the accuracy has greatly improved.(2)Non-stationary nonlinear multi-sequence mixed prediction model;Based on the nonlinear and non-stationary characteristics of the spatiotemporal sequence,a multi-sequence mixed prediction model M-WD-GRNN is established for the target sequence value,which is influenced by the historical data and the proximity time.Firstly,each monitoring sequence is denoised by wavelet,and then the multi-input and multi-output spatio-temporal sequence GRNN with spatial influence is established.The grid search and cross validation algorithm are introduced into the model Parameters are optimized.The experimental results show that the hybrid model proposed in this paper improves the prediction accuracy compared with other models,and also shows the advantages of wavelet transform in the prediction of groundwater depth.The model is used to predict the groundwater depth in the 12 th period.(3)Space-time interpolation simulation;A generalized regression neural network(GRNN)adaptive fitting time-space Kriging interpolation function is proposed to establish a generalized regression neural network space-time interpolation model for the problem that the covariance function of space-time Kriging interpolation is poor in versatility and low accuracy.GRNN-STK),the spatial distribution of groundwater depth in different areas of the study area was drawn.Through the analysis of the distribution map and the analysis of the specific influencing factors of the corresponding detection sites,the spatial pattern evolution of groundwater depth in the study area is revealed.The experimental results show that the model has a great improvement compared with the ordinary Kriging interpolation accuracy,and has made important prerequisite for establishing a reliable groundwater space forecasting system.For the relevant departments to take some efficient and reliable groundwater management and rational development of the decision-making,to provide adequate argument and reference.The innovation of this paper is that three new hybrid models are proposed by combining of non-linear method and theory to solve problem that we encountered in research,such as monitoring data missing,lack of the high efficiency and accuracy spatio-temporal multi-series prediction model and spatial distribution interpolation algorithm with lower accuracy and not universal et al..When take non-linear essence of groundwater level into account,we established a hybrid model for spatio-temporal data imputation,which is combination of classification algorithm,spatio-temporal correlation analysis,support vector machine and parameter optimization algorithm;Non-linear multi-series time domain hybrid predicted model is established by combining de-noising theory and generalized regression neural network;Spatio-temporal interpolation hybrid model is constructed based on generalized regression neural network and space-time kriging interpolation.Moreover,the accuracy and effectiveness of them have been tested by the data experiment.The work of this paper is that we explored the spatio-temporal simulation of the agent model,and can provide scientific method for quickly efficient to carry out spatial dynamic simulation and forecasting.