Local Prediction Model Research Articles

An improved chaos similarity model for hydrological forecasting

The local approximation prediction model (LPM) is a forecasting approach based on chaos theory. It uses the Euclidean distance to evaluate the spatial proximity between two phase points, which is a measure of “quantity” similarity. This study improved LPM by proposing a coupled “quantity-pattern” similarity model (CQPSM). The “pattern” similarity is defined as the similarity of two phase points on internal structure, and measured by a cumulative unit-step function. The forbearing stratified sequencing method is employed to solve the double-objective optimization problem. The proposed model was tested using the data from the Danjiangkou reservoir basin for future 12-monthly precipitation forecasting with lead times from a month to a year. Four performance measures (R, RMSE, MARE and NSE) were computed and compared to those of LPM and autoregressive (AR) model. The results show a better performance of CQPSM and indicates the importance of considering “pattern” similarity between two phase points. Moreover, the weighted Euclidean distance based on partial information logic was discussed and applied to further improve the “quantity” similarity. Consequently, the proposed CQPSM could provide a new alternative for hydrological forecasting.

Dynamic and Statistical Features of Internal Solitary Waves on the Continental Slope in the Northern South China Sea Derived From Mooring Observations

AbstractFor investigation of internal solitary waves (ISWs) in the South China Sea (SCS), most cruise observations are concentrated from Luzon Strait to Dongsha Atoll in the northeastern SCS but few on the continental slope far away from the west of Dongsha Atoll. In this study, we use 1‐year long mooring data to determine dynamic and statistical features of the ISWs on the shelf slope of the northwestern SCS. The analysis results of the mooring data reveal that the ocean internal waves on the shelf slope of the northwestern SCS have physical properties of highly nonlinear waves, which are well described by the solutions of the Korteweg‐de Vries equation. The mean nonlinear phase speeds of mode‐1 and mode‐2 ISWs are 1.38 ± 0.14 and 0.66 ± 0.12 m/s, respectively. The major direction of mode‐1 ISWs is northwestward 305° ± 21°. Strong ISW currents force the major direction of total current velocities to turn 67.5° in the upper layer and 135° in the lower layer. The monthly occurrence frequency distribution of ISWs shows a peak in July with a maximum frequency of 16.2% and a trough in March with a minimum frequency of 3.3%. Mode‐2 ISWs appear most in December, accounting for 50% of total mode‐2 ISWs. The largest mode‐2 ISW on record up today with the depressed amplitude as large as 91 m, and the elevated amplitude of 73 m was observed at mooring station. These new findings and new data are of significance to local internal wave prediction model development.

Decision Tree-PLS (DT-PLS) algorithm for the development of process: Specific local prediction models.

This work presents a novel multivariate statistical algorithm, Decision Tree-PLS (DT-PLS), to improve the prediction and understanding of dynamic processes based on local partial least square regression (PLSR) models for characteristic process groups defined based on Decision Tree (DT) analysis. The DT-PLS algorithm is successfully applied to two different cell culture data sets, one obtained from bioreactors of 3.5 L lab scale and the other obtained from the 15 ml ambr microbioreactor system. Substantial improvement in the predictive capabilities of the model can be achieved based on the localization compared to the classical PLSR approach, which is implemented in the commercially available packages. Additionally, the differences in the model parameters of the local models suggest that the governing process variables vary for the different process regimes indicating the different states of the cell under different process conditions.

Predictive Modeling of Pavement Damage Using Machine Learning and Big Data Processing

Potholes, soil settlement, and road subsidence have become major road safety hazards in South Korea. Such problems not only impede driver and pedestrian safety but also cause secondary accidents, economic losses, and damage the nation's image. To this end, we developed local predictive models that can be extrapolated to national estimation models. These models were developed from a specific area (Seoul Metropolitan City) that has the highest occurrences of potholes and road subsidence. This research utilized big data and artificial intelligence techniques to develop these models. The first step involved the dimensional reduction of independent variables using a mechanical-statistical approach. A data standardization process was then used for reducing the uncertainty of these variables. A total of 19 machine learning optimization methods were used to train the standardized variables. The optimized models were finally determined by an error comparison. As a result, the optimized prediction models for potholes, soil settlement, and road subsidence were found to be multiple regression analysis that showed an accuracy of 70% and robust regression analysis that showed an accuracy of 73%. Keywords: Pothole, Soil Settlement, Road Subsidence, Big Data, Machine Learning

Stochastic modeling of 3-D compositional distribution in the crust with Bayesian inference and application to geoneutrino observation in Japan

Geoneutrino observations, first achieved by KamLAND in 2005 and followed by Borexino in 2010, have accumulated statistics and improved sensitivity for more than ten years. The uncertainty of the geoneutrino flux at the surface is now reduced to a level small enough to set useful constraints on U and Th abundances in the Bulk Silicate Earth (BSE). However, in order to make inferences on earth’s compositional model, the contributions from the local crust need to be understood within a similar uncertainty. Here we develop a new method to construct a stochastic crustal composition model utilizing Bayesian inference. While the methodology has general applicability, it incorporates all the local uniqueness in its probabilistic framework. Unlike common approaches for this type of problem, our method does not depend on crustal segmentation into upper, (middle) and lower, whose classification and boundaries are not always well defined. We also develop a new modeling method to infer rock composition distributions that conserve mass balance and therefore do not bias the results. Combined with a new vast collection of geochemical data for rock samples in the Japan arc, we apply this method to geoneutrino observation at Kamioka, Japan. Currently a difficulty remains in the handling of correlations in the flux integration; we conservatively assume maximum correlation, which leads to large flux estimation errors of 60–70%. Despite the large errors, this is the first local crustal model for geoneutrino flux prediction with probabilistic error estimation in a reproducible way.

Varying dataset resolution alters predictive accuracy of spatially explicit ensemble models for avian species distribution.

Species distribution models can be made more accurate by use of new “Spatiotemporal Exploratory Models” (STEMs), a type of spatially explicit ensemble model (SEEM) developed at the continental scale that averages regional models pixel by pixel. Although SEEMs can generate more accurate predictions of species distributions, they are computationally expensive. We compared the accuracies of each model for 11 grassland bird species and examined whether they improve accuracy at a statewide scale for fine and coarse predictor resolutions. We used a combination of survey data and citizen science data for 11 grassland bird species in Oklahoma to test a spatially explicit ensemble model at a smaller scale for its effects on accuracy of current models. We found that only four species performed best with either a statewide model or SEEM; the most accurate model for the remaining seven species varied with data resolution and performance measure. Policy implications: Determination of nonheterogeneity may depend on the spatial resolution of the examined dataset. Managers should be cautious if any regional differences are expected when developing policy from range‐wide results that show a single model or timeframe. We recommend use of standard species distribution models or other types of nonspatially explicit ensemble models for local species prediction models. Further study is necessary to understand at what point SEEMs become necessary with varying dataset resolutions.

A novel model for hourly PM2.5 concentration prediction based on CART and EELM

Hourly PM2.5 concentrations have multiple change patterns. For hourly PM2.5 concentration prediction, it is beneficial to split the whole dataset into several subsets with similar properties and to train a local prediction model for each subset. However, the methods based on local models need to solve the global-local duality. In this study, a novel prediction model based on classification and regression tree (CART) and ensemble extreme learning machine (EELM) methods is developed to split the dataset into subsets in a hierarchical fashion and build a prediction model for each leaf. Firstly, CART is used to split the dataset by constructing a shallow hierarchical regression tree. Then at each node of the tree, EELM models are built using the training samples of the node, and hidden neuron numbers are selected to minimize validation errors respectively on the leaves of a sub-tree that takes the node as the root. Finally, for each leaf of the tree, a global and several local EELMs on the path from the root to the leaf are compared, and the one with the smallest validation error on the leaf is chosen. The meteorological data of Yancheng urban area and the air pollutant concentration data from City Monitoring Centre are used to evaluate the method developed. The experimental results demonstrate that the method developed addresses the global-local duality, having better performance than global models including random forest (RF), v-support vector regression (v-SVR) and EELM, and other local models based on season and k-means clustering. The new model has improved the capability of treating multiple change patterns.

Modeling of failure mode switching and shear band propagation using the correspondence framework of peridynamics

This study presents an approach to model dynamic failure mode switching and shear band propagation using the correspondence framework of state-based peridynamics. To effectively model spontaneous shear-band-to-crack switching phenomenon, which is of intrinsic complexity, a combined peridynamic nonlocal and classical local damage model for failure prediction is proposed. The concept of bond failure, often employed in the bond-based peridynamic modeling of brittle materials, is extended to the state-based peridynamics to model cleavage failure in elasto-viscoplastic materials so that the directional and progressive nature of damage can be readily handled. The classical constitutive relation for viscous fluid is adapted for use in non-ordinary state-based peridynamics to model stress collapsing state in shear bands. To show the effectiveness of the proposed approach, a comprehensive numerical study of a pre-notched plate subjected to asymmetric impact loading is conducted, including the phenomenon of shear-band-to-crack switching under an intermediate impact velocity and ductile failure under high impact velocity. A zero-energy mode suppression with an upper bound to avoid over-correction is introduced. By using the proposed modeling approach, dynamic failure mode switching and shear band propagation path are accurately captured.

Chaos-based multigene genetic programming: A new hybrid strategy for river flow forecasting

Chaos theory is integrated with Multi-Gene Genetic Programming (MGGP) engine as a new hybrid model for river flow forecasting. This is to be referred to as Chaos-MGGP and its performance is tested using daily historic flow time series at four gauging stations in two countries with a mix of both intermittent and perennial rivers. Three models are developed: (i) Local Prediction Model (LPM); (ii) standalone MGGP; and (iii) Chaos-MGGP, where the first two models serve as the benchmark for comparison purposes. The Phase-Space Reconstruction (PSR) parameters of delay time and embedding dimension form the dominant input signals derived from original time series using chaos theory and these are transferred to Chaos-MGGP. The paper develops a procedure to identify global optimum values of the PSR parameters for the construction of a regression-type prediction model to implement the Chaos-MGGP model. The inter-comparison of the results at the selected four gauging stations shows that the Chaos-MGGP model provides more accurate forecasts than those of stand-alone MGGP or LPM models.

Acoustic waveform inversion and uncertainty analysis for explosion yield estimation

Waveform inversion techniques are widely used to extract source information from acoustic signals. The methods exploit full-waveform information and provide further constraints on source inversion than other parameter-based inversions. Infrasound waveform inversion was recently applied to chemical explosions and showed promising results for the determination of explosion yield. The method can improve the accuracy of the estimated yield by using full 3-D numerical Green's functions and incorporating the propagation path effects into the inversion. However, the inversion results are often given without any uncertainty estimation, which is critical for a post-detonation forensic analysis. In this study, we evaluate the yield estimation errors associated with the uncertainty in the atmospheric representation. An ensemble of atmospheric realizations is obtained by adding random turbulence to an atmospheric model and the uncertainty of inversion solution depending on atmospheric variation is quantified. In addition, different atmospheric models derived from local weather data or numerical weather prediction models are validated for yield estimation by using a series of chemical explosion data.

The power load’s signal analysis and short-term prediction based on wavelet decomposition

The complex signal represented by power load is affected by many factors, so the signal components are very complicated. So that, it is difficult to obtain satisfactory prediction accuracy by using a single model for the complex signal. In this case, wavelet decomposition is used to decompose the power load into a series of sub signals. The low frequency sub signal is remarkably periodic, and the high frequency sub signals can prove to be chaotic signals. Then the signals of different characteristics are predicted by different models. For the low frequency sub signal, the support vector machine (SVM) is adopted. In SVM model, air temperature and week attributes are included in model inputs. Especially the week attribute is represented by a 3-bit binary encoding, which represents Monday to Sunday. For the chaotic high frequency sub signals, the chaotic local prediction (CLP) model is adopted. In CLP model, the embedding dimension and time delay are key parameters, which determines the prediction accuracy. In order to find the optimal parameters, a segmentation validation algorithm is proposed in this paper. The algorithm segments the known power load according to the time sequence. Then, based on the segmentation data, the optimal parameters are chosen based on the prediction accuracy. Compared with a single model, the prediction accuracy of the proposed algorithm is improved obviously, which proves the effectiveness.

Active Learning with Multiple Localized Regression Models

Oftentimes businesses face the challenge of requiring costly information to improve the accuracy of prediction tasks. One notable example is obtaining informative customer feedback (e.g., customer-product ratings via costly incentives) to improve the effectiveness of recommender systems. In this paper, we develop a novel active learning approach, which aims to intelligently select informative training instances to be labeled so as to maximally improve the prediction accuracy of a real-valued prediction model. We focus on large, heterogeneous, and dyadic data, and on localized modeling techniques, which have been shown to model such data particularly well, as compared to a single, “global” model. Importantly, dyadic data with covariates is pervasive in contemporary big data applications such as large-scale recommender systems and search advertising. A key benefit from incorporating dyadic information is their simple, meaningful representation of heterogeneous data, in contrast to alternative local modeling techniques that typically produce complex and incomprehensible predictive patterns. We develop a computationally efficient active learning policy specifically tailored to exploit multiple local prediction models to identify informative acquisitions. Existing active learning policies are often computationally prohibitive for the setting we explore, and our policy makes the application of active learning computationally feasible for this setting. We present comprehensive empirical evaluations that demonstrate the benefits of our approach and explore its performance in real world, challenging domains.

The influence of abiotic and biotic factors on the invasion of Ixodes scapularis in Ontario, Canada

In northeastern North America, the blacklegged tick, Ixodes scapularis, is the vector of numerous tick-borne pathogens, including the agent of Lyme disease, Borrelia burgdorferi sensu stricto. Since 1990, there has been a rapid spread of I. scapularis northward into the province of Ontario, Canada. Climate change has been implicated as one of the driving factors for the spread of this vector. Other ecological factors also influence survival of I. scapularis populations and may facilitate invasion. The objective of this study was to identify local abiotic and biotic factors of significance for the invasion of I. scapularis in Ontario. The presence of ticks was determined by drag sampling at 154 sites in southern, eastern and central Ontario from May to October in 2014 and 2015. At each site, data on site aspect, forest cover, understory density and composition, soil moisture and composition, and the depth of litter layer were collected. Cumulative degree days above zero °C, total precipitation and elevation were attributed to each site using a geographic information system. A mixed multi-variable logistic regression model was created to assess the impact of the ecological factors on the presence of I. scapularis. In total, I. scapularis was found at 29 sites (18.8%) across the study area. The density of the understory, the presence of shrubs and the interaction of these two ecological factors were statistically significant, as well as longitude and cumulative degree days above zero. Our findings illustrate that local ecological factors are of importance for the invasion of I. scapularis into Ontario, and may be used to enhance local public health interventions and current predictive models and risk maps for I. scapularis.

Avaliação da dependência espacial na modelagem do desempenho da segurança viária em zonas de tráfego

Uma técnica comumente usada no processo de modelagem do Desempenho da Segurança Viária (DSV), no nível de planejamento, são os Modelos Lineares Generalizados (MLG) assumindo a distribuição binomial negativa para os erros. Limitações dessa técnica, por não considerar os efeitos espaciais, têm sido contornadas com a utilização de modelos espaciais locais a partir de técnicas de regressão espacial como a Regressão de Poisson Geograficamente Ponderada (RPGP). Este trabalho tem por objetivo apresentar uma análise comparativa entre modelos de previsão de acidentes globais não espaciais e locais espaciais para a estimação do DSV agregado ao nível de zonas de tráfego em Fortaleza/CE. Foram calibrados modelos para a variável dependente acidentes totais e acidentes com vítimas e os resultados mostraram que os modelos RPGP apre-sentaram melhor desempenho que os MLG nas medidas de ajustes e na redução da autocorrelação espacial dos resíduos, sendo capazes de captar a heterogeneidade espacial da frequência dos acidentes.

Validation of Long-Term Pavement Performance Prediction Models for Resilient Modulus of Unbound Granular Materials

Reliable characterization of unbound granular material and subgrade resilient modulus requires repeated load triaxial testing, advanced dynamic loading equipment, and technical experience not typically available in many geotechnical laboratories. For Level 2 design inputs, the Mechanistic–Empirical Pavement Design Guide recommends correlation models to estimate the resilient modulus of unbound granular material and subgrade from basic material properties (e.g., gradation, unconfined compressive strength, and California bearing ratio). These correlation models are developed from data on a wide range of soil and material types; this wide range increases the associated error with resilient modulus prediction. This paper compares laboratory measured resilient modulus for two types of locally available unbound granular materials to the predicted values; models developed under the Long-Term Pavement Performance (LTPP) program were used. Resilient modulus tests of six gradations of two types of material—100% crushed limestone and gravel—were conducted at two levels of moisture content. Results showed that the LTPP models significantly underestimated resilient modulus for the limestone material and showed high residuals associated with evaluating resilient modulus of both gravel and limestone materials. The coefficient of variation of the root mean square error was 50.3% for gravel and 55.6% for limestone. The high residuals associated with evaluating resilient modulus with the LTPP models justify the development of local prediction models to improve the reliability of Level 2 design inputs for unbound granular materials.

Development of Traffic Accident frequency Prediction Model by Administrative zone - A Case of Seoul

In Korea, the local traffic safety master plan has been established and implemented according to the Traffic Safety Act. Each local government is required to establish a customized traffic safety policy and share roles for improvement of traffic safety and this means that local governments lead and promote effective local traffic safety policies fit for local circumstances in substance. For implementing efficient traffic safety policies, which accord with many-sided characteristics of local governments, the prediction of community-based traffic accidents, which considers local characteristics and the analysis of accident influence factors must be preceded, but there is a shortage of research on this. Most of existing studies on the community-based traffic accident prediction used social and economic variables related to accident exposure environments in countries or cities due to the limit of collected data. For this reason, there was a limit in applying the developed models to the actual reduction of traffic accidents. Thus, this study developed a local traffic accident prediction model, based on smaller regional units, administrative districts, which were not omitted in existing studies and suggested a method to reflect traffic safety facility and policy variables that traffic safety policy makers can control, in addition to social and economic variables related to accident exposure environments, in the model and apply them to the development of local traffic safety policies. The model development result showed that in terms of accident exposure environments, road extension, gross floor area of buildings, the ratio of bus lane installation and the number of crossroads and crosswalks had a positive relation with accidents and the ratio of crosswalk sign installation, the number of speed bumps and the results of clampdown by police force had a negative relation with accidents.

Evaluating the dual‐boundary forcing concept in subsurface–land surface interactions of the hydrological cycle

AbstractSubsurface and land surface processes (e.g. groundwater flow, evapotranspiration) of the hydrological cycle are connected via complex feedback mechanisms, which are difficult to analyze and quantify. In this study, the dual‐boundary forcing concept that reveals space–time coherence between groundwater dynamics and land surface processes is evaluated. The underlying hypothesis is that a simplified representation of groundwater dynamics may alter the variability of land surface processes, which may eventually affect the prognostic capability of a numerical model. A coupled subsurface–land surface model ParFlow.CLM is applied over the Rur catchment, Germany, and the mass and energy fluxes of the coupled water and energy cycles are simulated over three consecutive years considering three different lower boundary conditions (dynamic, constant, and free‐drainage) based on groundwater dynamics to substantiate the aforementioned hypothesis. Continuous wavelet transform technique is applied to analyze scale‐dependent variability of the simulated mass and energy fluxes. The results show clear differences in temporal variability of latent heat flux simulated by the model configurations with different lower boundary conditions at monthly to multi‐month time scales (~32–91 days) especially under soil moisture limited conditions. The results also suggest that temporal variability of latent heat flux is affected at even smaller time scales (~1–3 days) if a simple gravity drainage lower boundary condition is considered in the coupled model. This study demonstrates the importance of a physically consistent representation of groundwater dynamics in a numerical model, which may be important to consider in local weather prediction models and water resources assessments, e.g. drought prediction. Copyright © 2015 John Wiley & Sons, Ltd.

Development of an Integrated Forecasting and Warning System for Abrupt Natural Disaster using rainfall prediction data and Ubiquitous Sensor Network(USN)

The objectives of this research have been focussed on 1) developing prediction techniques for the flash flood and landslide based on rainfall prediction data in agricultural area and 2) developing an integrated forecasting system for the abrupt disasters using USN based real-time disaster sensing techniques. This study contains following steps to achieve the objective; 1) selecting rainfall prediction data, 2) constructing prediction techniques for flash flood and landslide, 3) developing USN and communication network protocol for detecting the abrupt disaster suitable for rural area, & 4) developing mobile application and SMS based early warning service system for local resident and tourist. Local prediction model (LDAPS, UM1.5km) supported by Korean meteorological administration was used for the rainfall prediction by considering spatial and temporal resolution. NRCS TR-20 and infinite slope stability analysis model were used to predict flash flood and landslide. There are limitations in terms of communication distance and cost using Zigbee and CDMA which have been used for existing disaster sensors. Rural suitable sensor-network module for water level and tilting gauge and gateway based on proprietary RF network were developed by consideration of low-cost, low-power, and long-distance for communication suitable for rural condition. SMS & mobile application forecasting & alarming system for local resident and tourist was set up for minimizing damage on the critical regions for abrupt disaster. The developed H/W & S/W for integrated abrupt disaster forecasting & alarming system was verified by field application.

Wind speed forecast correction models using polynomial neural networks

Accurate short-term wind speed forecasting is important for the planning of a renewable energy power generation and utilization, especially in grid systems. In meteorology it is usual to improve the forecasts by means of some post-processing methods using local measurements and weather prediction model outputs. Neural networks, trained with local real data observations can improve short-term wind speed forecasts with respect to meso-scale numerical meteorological model outcomes of the same data types in the majority of cases. Large-scale forecast models are based on the numerical integration of differential equation systems, which can describe atmospheric circulation processes on account of global meteorological observations. Several layer 3D complex models, which involve a large number of matrix variables, cannot exactly describe conditions near the ground, highly influenced by a landscape relief, coast, structure and other factors. Polynomial neural networks can form and solve general differential equations, which allow to model real complex systems by means of substitution derivative term sum series. The proposed adaptive method forms a correction function according to real observations and consequently applies forecasts to revise a desired prognosis in a selected locality.

Prediction of soil organic matter using a spatially constrained local partial least squares regression and the Chinese vis–NIR spectral library

SummaryWe need to determine the best use of soil vis–NIR spectral libraries that are being developed at regional, national and global scales to predict soil properties from new spectral readings. To reduce the complexity of a calibration dataset derived from the Chinese vis–NIR soil spectral library (CSSL), we tested a local regression method that combined geographical sub‐setting with a local partial least squares regression (local‐PLSR) that uses a limited number of similar vis–NIR spectra (k‐nearest neighbours). The central idea of the local regression, and of other local statistical approaches, is to derive a local prediction model by identifying samples in the calibration dataset that are similar, in spectral variable space, to the samples used for prediction. Here, to derive our local regressions we used Euclidean distance in spectral space between the calibration dataset and prediction samples, and we also used soil geographical zoning to account for similarities in soil‐forming conditions. We tested this approach with the CSSL, which comprised 2732 soil samples collected from 20 provinces in the People's Republic of China to predict soil organic matter (SOM). Results showed that the prediction accuracy of our spatially constrained local‐PLSR method (R2 = 0.74, RPIQ = 2.6) was better than that from local‐PLSR (R2 = 0.69, RPIQ = 2.3) and PLSR alone (R2 = 0.50, RPIQ = 1.5). The coupling of a local‐PLSR regression with soil geographical zoning can improve the accuracy of local SOM predictions using large, complex soil spectral libraries. The approach might be embedded into vis–NIR sensors for laboratory analysis or field estimation.