Multi-model Combination Research Articles

Exploration of virtual city construction and optimization based on deep learning

With continuous artificial intelligence and computer graphics technology, virtual cities are receiving widespread attention as an essential digital twin technology, . The core issue of this study is how to choose appropriate neural networks and algorithms to build models to construct virtual cities. The research methods include literature search, research and improvement of deep learning algorithms, and exploration of multi-model combinations. The research conclusion shows that choosing appropriate neural networks and algorithms is the key to building high-quality virtual cities, and targeted improvement and optimization of deep learning algorithms can further improve the accuracy and efficiency of virtual city construction. The strategy of multi-model combination also shows its unique advantages. By integrating different neural networks and algorithms, people can fully utilize their advantages and compensate for each other's deficiencies. With the advancement of technology, more innovative methods and technologies will be applied to this, which will help to build a more realistic virtual world and promote the development and application of virtual cities.

Predicting maximum scour depth uncertainty downstream of grade control structures: a multi-model approach with Bayesian model averaging

This research proposes a novel multi-model combination approach to overcome the limitations of traditional methods in predicting the scour depth downstream of grade control structures. The study evaluates 26 scenarios using a combination of five individual models and assesses the efficiency of four combination methods. Results indicate that GEP and SVM outperform traditional combination methods. Additionally, the study employs Bayesian Model Average (BMA) to assess the uncertainty originating from individual models. The research concludes by providing predictive equations for GEP and SVM, enabling users to apply these methods for accurate maximum scour depth prediction. In the ideal case, the SVM model had an R 2 = 0.91, with an RMSE of 0.84, whereas the GEP model achieved an R 2 = 0.94, with an RMSE of 0.67. The proposed multi-model approach is more reliable and accurate for predicting maximum scour depth downstream of GCS.

Multi-model combination in key steps for landslide susceptibility modeling and uncertainty analysis: a case study in Baoji City, China

Reliable landslide susceptibility maps are essential for geohazard risk management. However, the selection of models and methods in the landslide susceptibility modeling (LSM) process is subjective and can lead to uncertainties in susceptibility outcomes. This paper introduces a framework for LSM and uncertainty analysis that leverages multiple models in key steps. The framework dynamically operates and combines models at each LSM step to generate susceptibility evaluation results from various method combinations. The uncertainty of different model combinations is then analyzed by comparing these results. The framework’s effectiveness is validated using Baoji City as a case study, examining the impact of different attribute interval numbers (AIN), non-landslide negative sample selection methods, and prediction models on susceptibility prediction outcomes. The results of each group show that the highest AUC value is 0.963 (AIN = 12, buffer-controlled sampling, Random forest) is about 0.26 higher than the lowest (no AIN, low-slope controlled sampling, Support vector machine). The findings suggest that the combinations involving larger AIN values, buffer-controlled sampling, XGBoost, or Random forest model yield relatively high AUC accuracy and relatively low uncertainty in the susceptibility index.

Improving performance of bucket-type hydrological models in high latitudes with multi-model combination methods: Can we wring water from a stone?

Multi-model combination (averaging) methods (MMCMs) are used to improve the accuracy of hydrological (precipitation-runoff) outputs in simulation or forecasting/prediction modes. In this paper, we examined if the application of MMCMs can improve model performance in reproducing distributions of hydrological signatures, such as annual maxima or minima of varying durations. To this end, ten MMCMs were applied to 29 bucket-type models to simulate runoff in 50 high-latitude catchments. The MMCMs were evaluated by comparing the resulting simulated flows to the reference (i.e., best-performing) individual model, considering various commonly used performance indicators, as well as model performance in reproducing the distributions of signatures. Additionally, we analysed whether (1) the selection of the candidate models, or (2) targeting specific signatures, such as annual maxima or minima, can improve performance of the model combinations. The results suggest that the application of MMCMs can improve accuracy of runoff simulations in terms of traditional performance indicators, but fails to improve performance in reproducing the distributions of signatures. Neither excluding poor-performing models nor applying the MMCMs with the targeted signatures, improves this aspect of model performance. These findings clearly reveal the need for further research aiming at enhancing model performance in reproducing the distributions of hydrological signatures, which is essential for climate-change impact studies.

Jumpiness in Ensemble Forecasts of Atlantic Tropical Cyclone Tracks

Abstract We investigate the run-to-run consistency (jumpiness) of ensemble forecasts of tropical cyclone tracks from three global centers: ECMWF, the Met Office, and NCEP. We use a divergence function to quantify the change in cross-track position between consecutive ensemble forecasts initialized at 12-h intervals. Results for the 2019–21 North Atlantic hurricane season show that the jumpiness varied substantially between cases and centers, with no common cause across the different ensemble systems. Recent upgrades to the Met Office and NCEP ensembles reduced their overall jumpiness to match that of the ECMWF ensemble. The average divergence over the set of cases provides an objective measure of the expected change in cross-track position from one forecast to the next. For example, a user should expect on average that the ensemble mean position will change by around 80–90 km in the cross-track direction between a forecast for 120 h ahead and the updated forecast made 12 h later for the same valid time. This quantitative information can support users’ decision-making, for example, in deciding whether to act now or wait for the next forecast. We did not find any link between jumpiness and skill, indicating that users should not rely on the consistency between successive forecasts as a measure of confidence. Instead, we suggest that users should use ensemble spread and probabilistic information to assess forecast uncertainty, and consider multimodel combinations to reduce the effects of jumpiness. Significance Statement Forecasting the tracks of tropical cyclones is essential to mitigate their impacts on society. Numerical weather prediction models provide valuable guidance, but occasionally there is a large jump in the predicted track from one run to the next. This jumpiness complicates the creation and communication of consistent forecast advisories and early warnings. In this work we aim to better understand forecast jumpiness and we provide practical information to forecasters to help them better use the model guidance. We show that the jumpiest cases are different for different modeling centers, that recent model upgrades have reduced forecast jumpiness, and that there is not a strong link between jumpiness and forecast skill.

Mine Surface Settlement Prediction Based on Optimized VMD and Multi-Model Combination

The accurate prediction of mining area surface deformation is essential to preventing large-scale coal mining-related surface collapse and ensure safety and daily life continuity. Monitoring subsidence in mining areas is challenged by environmental interference, causing data noise. This paper employs the Sparrow Search Algorithm, which integrates Sine Cosine and Cauchy mutation (SCSSA), to optimize variational mode decomposition (VMD) and combine multi-models for prediction. Firstly, SCSSA is employed to adaptively determine the parameters of VMD using envelope entropy as the fitness value. Subsequently, the VMD method optimized using SCSSA adaptively decomposes the original mining area subsidence data sequence into various sub-sequences. Then, SCSSA-VMD is applied to adaptively decompose the original mining subsidence data sequence into multiple sub-sequences. Meanwhile, using sample entropy, the sub-sequences are categorized into trend sequences and fluctuation sequences, and different models are employed to predict sub-sequences at different frequencies. Finally, the prediction results from different sub-sequences are integrated to obtain the final prediction of mining area subsidence. To validate the predictive performance of the established model, experiments are conducted using GNSS monitoring data from the 110801 working face of Banji Coal Mine in Bozhou. The results demonstrate the following: (1) The hybrid model enhanced the prediction accuracy and trends by decomposing the data and optimizing the parameters with VMD. It outperformed single models, reducing errors and improving predictive trends. (2) The hybrid model significantly improved the prediction accuracy for subsidence data at work surface monitoring stations. It is particularly effective at critical subsidence points, making it a valuable reference for safety in mining operations.

Short-term passenger flow prediction based on deep learning

With the continuous improvement of urbanization, the problem of urban congestion has become increasingly prominent. Rail transit can greatly alleviate the congestion of ground traffic and is considered as a feasible solution to alleviate urban congestion. However, there are many factors affecting rail transit, among which the number of passengers accounts for the majority. Excessive passenger flow will sharply increase the probability of public safety events in confined space. The continuous development of deep learning has benefited from studies such as convolutional neural networks (CNN), so as to correctly predict short-term traffic. Despite continual model accuracy improvement, prior research has given little regard to the application boundaries of diverse networks in traffic prediction. In this paper, by comparing the results of previous studies, the prediction effect and deficiency of long short-term memory (LSTM), CNN, graph convolutional network (GCN) are discussed. And results reveal that prediction models relying on recurrent neural networks (RNN), such as LSTM, can capture the time dimension of short-term passengers very well, but not the spatial patterns, and model training time is lengthy. The GCN model would account for the spatial dependence of the traffic network, but it has poor performance when constructing the deep network. In general, each of these three models has its own unique aspects. Due to some problems in a single model, the multi-model combination deep learning framework is gradually emerging. In practical application, the complexity of the model, performance effect and application value should be considered comprehensively.

Adaptive selection and optimal combination scheme of candidate models for real-time integrated prediction of urban flood

The ability to predict urban floods is crucial for reducing potential losses. Previous studies suggest that a multimodel combination is an effective way to improve the prediction performance of urban flood models; however, few studies have systematically investigated the impact of candidate models on the performance of the integrated model. Therefore, this study proposes a multimodel integrated forecasting method for urban flooding from the perspective of the response relationship between the candidate models and integrated model. The results of this study suggest that the prediction error of the proposed was reduced by 46.9%–64.6% compared with that of the single model. The results of various candidate model combinations indicate that there is a threshold effect for the number of candidate models in the integrated model; the integrated model with six candidate models exhibited the highest prediction accuracy. However, the increase in the number of candidate models was accompanied by a significant decrease in computational efficiency of the integrated model. Based on the accuracy and timeliness requirements of urban flood prediction, a scheme combining gradient lifting decision tree, random forest, back propagation, and support vector machine models was found to be the best candidate model combination scheme. The real-time warning results of the aforementioned combination model provided superior warning performance. The results of this study provide a reference for the construction of more suitable urban flood models, real-time forecasting, and warnings.

Pose estimation for swimmers in video surveillance

Traditional models for pose estimation in video surveillance are based on graph structures, in this paper, we propose a method that breaks the limitation of template matching within a range of pose changes to obtain robust results. We implement our swimmer pose estimation method based on deep learning. We take use of High-Resolution Net (HRNet) to extract and fuse visual features of visual object and complete the object detection using the key points of human joint. The proposed model could be applied to all kinds of swimming styles throughout appropriate training. Compared with the methods that require multimodel combinations and training, the proposed method directly achieves the end-to-end prediction, which is easily to be implemented and deployed. In addition, a cross-fusion module is added between parallel networks, which assists the network to make use of the characteristics of multiple resolutions. The proposed network has achieved ideal results in the pose estimation of swimmers by comparing HRNet-W32 and HRNet-W48. In addition, we propose an annotated key point dataset of swimmers which was created from the view of underwater swimmers. Compared with side view, the torso of swimmers collected by the underwater view is much suitable for a broad spectrum of machine vision tasks.

Application of hydrological models in climate change framework for a river basin in India

Abstract Soil Water Assessment Tool (SWAT), Hydrologic Engineering Center-Hydrologic Modelling System (HEC-HMS), and Hydrologic Simulation Program Fortran (HSPF) are explored for streamflow simulation of Lower Godavari Basin, India. The simulating ability of models is evaluated using four indicators. SWAT has shown exceptional simulating ability in calibration and validation compared to the other two. Accordingly, SWAT is used in the climate change framework using an ensemble of 13 Global Climate Models and four Shared Socioeconomic Pathways (SSPs). Three time segments, near-future (2021–2046), mid-future (2047–2072), and far-future (2073–2099), are considered for analysis. Four SSPs show a substantial increase in streamflow compared to the historical period (1982–2020). These deviations range from 17.14 (in SSP245) to 28.35% (in SSP126) (near-future), 31.32 (SSP370) to 43.28% (SSP585) (mid-future), and 30.41 (SSP126) to 70.8% (SSP585) (far-future). Across all timescales covering 948 months, the highest projected streamflows observed in SSP126, SSP245, SSP370, and SSP585 were 4962.36, 6,108, 6,821, and 6,845 m3/s, respectively. Efforts are also made to appraise the influence of multi-model combinations on streamflow. The present study is expected to provide a platform for holistic decision-making, which helps develop efficient basin planning and management alternatives.

Short-Term Subway Passenger Flow Prediction Based on Time Series Adaptive Decomposition and Multi-Model Combination (IVMD-SE-MSSA)

The accurate forecasting of short-term subway passenger flow is beneficial for promoting operational efficiency and passenger satisfaction. However, the nonlinearity and nonstationarity of passenger flow time series bring challenges to short-term passenger flow prediction. To solve this challenge, a prediction model based on improved variational mode decomposition (IVMD) and multi-model combination is proposed. Firstly, the mixed-strategy improved sparrow search algorithm (MSSA) is used to adaptively determine the parameters of the VMD with envelope entropy as the fitness value. Then, IVMD is applied to decompose the original passenger flow time series into several sub-series adaptively. Meanwhile, the sample entropy is utilized to divide the sub-series into high-frequency and low-frequency components, and different models are established to predict the sub-series with different frequencies. Finally, the MSSA is employed to determine the weight coefficients of each sub-series to combine the prediction results of the sub-series and get the final passenger flow prediction results. To verify the prediction performance of the established model, passenger flow datasets from four different types of Nanning Metro stations were taken as examples for carrying out experiments. The experimental results showed that: (a) The proposed hybrid model for short-term passenger flow prediction is superior to several baseline models in terms of both prediction accuracy and versatility. (b) The proposed hybrid model is excellent in multi-step prediction. Taking station 1 as an example, the MAEs of the proposed model are 3.677, 5.7697, and 8.1881, respectively, which can provide technical support for subway operations management.

A Power System Timing Data Recovery Method Based on Improved VMD and Attention Mechanism Bi-Directional CNN-GRU

The temporal data of the power system are expanding with the growth of the power system and the proliferation of automated equipment. However, data loss may arise during the acquisition, measurement, transmission, and storage of temporal data. To address the insufficiency of temporal data in the power system, this study proposes a sequence-to-sequence (Seq2Seq) architecture to restore power system temporal data. This architecture comprises a radial convolutional neural unit (CNN) network and a gated recurrent unit (GRU) network. Specifically, to account for the periodicity and volatility of temporal data, VMD is employed to decompose the time series data output into components of different frequencies. CNN is utilized to extract the spatial characteristics of temporal data. At the same time, Seq2Seq is employed to reconstruct each component based on introducing a feature timing and multi-model combination triple attention mechanism. The feature attention mechanism calculates the contribution rate of each feature quantity and independently mines the correlation between the time series data output and each feature value. The temporal attention mechanism autonomously extracts historical–critical moment information. A multi-model combination attention mechanism is introduced, and the missing data repair value is obtained after modeling the combination of data on both sides of the missing data. Recovery experiments are conducted based on actual data, and the method’s effectiveness is verified by comparison with other methods.

A reduced-form ensemble of short-term air quality forecasting with the Sparrow search algorithm and decomposition error correction.

The level of air pollution is reflected by the air quality index (AQI). People can use the AQI to organize their activities in a way that reduces or prevents exposure to air pollution altogether. Based on the AQI, governments, organizations, and businesses can also make plans to reduce air pollution. The multi-model ensemble has recently become a popular method for forecasting time series; however, it encounters the research problems of multi-parameter optimization and interaction analysis. To this end, a reduced-form ensemble of short-term air quality forecasting with the Sparrow search algorithm and decomposition error correction model is proposed in this paper. First, the data are decomposed using the CEEMDAN decomposition algorithm. Second, the Sparrow search algorithm is used in the model training process to obtain the optimal hyperparameters of the deep learning model and construct the optimal deep learning model. Next, the constructed models are used to predict the decomposed data, and the Lagrange multiplier method is used to determine the weights of each deep learning model. At last, the prediction results of each deep learning model are combined according to the weights to obtain the combined prediction results. Experiments show that (1) GRU, Bi-GRU, LSTM, and Bi-LSTM are used to predict the undecomposed data and the data decomposed by CEEMADN. The outcomes demonstrate that the CEEMDAN decomposition technique can enhance the accuracy of the forecast, specifically an 11.248% reduction in average RMSE and a 0.865% increase in average R2. (2) A multi-model combination method based on the Lagrange multiplier method is designed, which can obtain the weights of each deep learning model, and the weights can combine multiple models. The results of the multi-model combination are better than those of the single model. (3) The Lagrange multiplier method was compared with the simple average combination model and the MAE inverse combination model. The experimental results show that the results obtained using the Lagrange multiplier method are better than the other two.

Logistic Forecast Analysis of Sichuan Province on the Basis of Multi-model Combination

Sichuan province serving as the hub of comprehensive transportation in the southwest of our country, has the longest transportation lines in China and its freight transportation maintains a healthy development. In order to give a quantitative reference for the government to formulate logistics improvement and development policies, determining the scale of logistics infrastructure construction and analyzing the situation of the logistics market, this paper establishes several forecasting models to reveal the mechanism and forecast the logistics demand. Freight turnover is used as an indicator to reveal logistics demand and seven variables are summarized as influential factors. Then, multiple regression model, BP neural network model and grey forecasting model are available ways to do the forecast and their forecasting error is measured by several criterions. In order to avoid the technical shortage of a single model, this paper give each of the model a weight ratio and use a combination model to analyze the logistics demand of Sichuan Province. Since the multi-model has low overall error rate, using this combination model to predict the logistics needs of Sichuan in the coming years is pretty feasible.

Outcomes of the WMO Prize Challenge to Improve Subseasonal to Seasonal Predictions Using Artificial Intelligence

Abstract There is a high demand and expectation for subseasonal to seasonal (S2S) prediction, which provides forecasts beyond 2 weeks, but less than 3 months ahead. To assess the potential benefit of artificial intelligence (AI) methods for S2S prediction through better postprocessing of ensemble prediction system outputs, the World Meteorological Organization (WMO) coordinated a prize challenge in 2021 to improve subseasonal prediction. The goal of this competition was to produce the most skillful forecasts of precipitation and 2-m temperature globally averaged over forecast weeks 3 and 4 and over weeks 5 and 6 for the year 2020 using artificial intelligence techniques. The top three submissions, described in this article, succeeded in producing S2S forecasts significantly more skillful than the bias-corrected ECMWF operational reference forecasts, particularly for precipitation, through improved calibration of the ECMWF raw forecast outputs or multimodel combination. These forecast improvements should benefit the use of S2S forecasts in applications.

E3W—A Combined Model Based on GreedySoup Weighting Strategy for Chinese Agricultural News Classification

With the continuous development of the internet and big data, modernization and informatization are rapidly being realized in the agricultural field. In this line, the volume of agricultural news is also increasing. This explosion of agricultural news has made accurate access to agricultural news difficult, and the spread of news about some agricultural technologies has slowed down, resulting in certain hindrance to the development of agriculture. To address this problem, we apply NLP to agricultural news texts to classify the agricultural news, in order to ultimately improve the efficiency of agricultural news dissemination. We propose a classification model based on ERNIE + DPCNN, ERNIE, EGC, and Word2Vec + TextCNN as sub-models for Chinese short-agriculture text classification (E3W), utilizing the GreedySoup weighting strategy and multi-model combination; specifically, E3W consists of four sub-models, the output of which is processed using the GreedySoup weighting strategy. In the E3W model, we divide the classification process into two steps: in the first step, the text is passed through the four independent sub-models to obtain an initial classification result given by each sub-model; in the second step, the model considers the relationship between the initial classification result and the sub-models, and assigns weights to this initial classification result. The final category with the highest weight is used as the output of E3W. To fully evaluate the effectiveness of the E3W model, the accuracy, precision, recall, and F1-score are used as evaluation metrics in this paper. We conduct multiple sets of comparative experiments on a self-constructed agricultural data set, comparing E3W and its sub-models, as well as performing ablation experiments. The results demonstrate that the E3W model can improve the average accuracy by 1.02%, the average precision by 1.62%, the average recall by 1.21%, and the average F1-score by 1.02%. Overall, E3W can achieve state-of-the-art performance in Chinese agricultural news classification.

A stacking-based short-term wind power forecasting method by CBLSTM and ensemble learning

Aiming at the problem that the traditional wind power forecasting is difficult to deal with a large amount of strong volatility data and limited processing capacity of time series, a wind power forecasting method based on multi-model combination under stacking framework was proposed. First, the wind turbine data are cleaned by density-based spatial clustering of applications with the noise clustering method. Considering the differences of data observation and training principles, the proposed stacking method embedded multiple machine learning algorithms to utilize their diversified strength. The stacking base-learner includes the CBLSTM model, which has the advantages of deep architecture feature extraction, and takes into account data timing and nonlinear relationship as well as XGBoost and other tree ensemble learning models that were suitable for complex data modeling. The feasibility of the algorithm was verified by using the actual wind power data of two wind farms in Northeast and Western China. Experimental results show that the stacking ensemble learning method proposed has better forecasting performance and stability than other single forecasting models, which is of great significance to guide wind power dispatching operation and improve wind power consumption capacity.

Research and application of simulation system for reduction and stabilization process of chromium contaminated soil based on intelligent model

Reducing the impact of chromium-contaminated soil on humans and organisms is a worldwide problem, and it is urgent to treat and evaluate it. Stabilization efficiency is a key factor for screening and evaluating reduction stabilization process of chromium contaminated soil. In this study, various data-driven models were constructed to predict the reduction and stabilization efficiency of chromium contaminated soil, which has very important practical significance. The results show that both models can predict the stability efficiency, and the model constructed by multi-model combination modeling method has better prediction performance than the model established by single neural network. The prediction effect of multi-model nonlinear combination modeling method is better than that of linear combination modeling method, and the nonlinear combination model based on BPNN combiner combination SVM sub-model has the best performance. Our research provides a good idea for the modeling and prediction of stabilization efficiency, which greatly reduces the time and economic cost required for the reduction and stabilization process of chromium contaminated soil.

Performance dependence of multi-model combination methods on hydrological model calibration strategy and ensemble size

The multi-model combination is a technique to improve the performances of hydrological streamflow simulations. An area that has not been investigated much is the performance dependence of combination techniques on the hydrological model calibration strategy and ensemble size. This study aims at investigating the joint effect of the hydrological models, calibration strategies and ensemble sizes on combination abilities for selecting the most appropriate multi-model combination method. The ensemble members were constructed by applying four hydrological models and four objective functions over 383 catchments in China. The ensemble members were combined by using nine commonly used methods, which are Equal Weights (EWA), Akaike Information Criterion (AICA), Bayes Information Criterion (BICA), Bates and Granger (BGA), Granger Ramanathan A, B, and C (GRA, GRB, and GRC), Bayesian Model Averaging (BMA) and Multi-model Super Ensemble (MMSE). The GRC is found as the best multi-model combination method for hydrological simulations. Adding ensemble members by either multiple hydrological models or calibration strategies could help to improve the simulation abilities. Specifically, the increase of ensemble members can obviously enhance the performance of multi-model combinations when the ensemble size is less than six, while only limited improvement is achieved when the ensemble size is more than nine. The combination of ensemble members with various calibration strategies is hard to compensate for the weakness of hydrological model structures. As well, the application of a single calibration strategy in ensemble members only emphasizes single discharge periods and neglects other important discharge periods. This study found that various models with different objective functions are more robust and efficient. The combination performs better than any individual model in terms of Nash–Sutcliffe efficiency (NSE) for approximately 70% catchments, but the multi-model combination is less efficient in terms of low-flow simulations.

A new approach to extended‐range multimodel forecasting: Sequential learning algorithms

AbstractMultimodel combinations are a well‐established methodology in weather and climate prediction and their benefits have been widely discussed in the literature. Typical approaches involve combining the output of different numerical weather prediction (NWP) models using constant weighting factors, either uniformly distributed or determined through a prior skill assessment. This strategy, however, can lead to suboptimal levels of skill, as the performance of NWP models can vary with time (e.g., seasonally varying skill, changes in the forecasting system). Moreover, standard combination methods are not designed to incorporate predictions derived from sources other than NWP systems (e.g., climatological or time‐series forecasts). New algorithms developed within the machine learning community provide the opportunity for “online prediction” (also referred to as “sequential learning”). These methods consider a set of weighted predictors or “experts” to produce subsequent predictions in which the combination or “mixture” is updated at each step to optimize a loss or skill function. The predictors are highly flexible and can combine both NWP and statistically derived forecasts transparently. A set of these online prediction methods is tested and compared with standard multimodel combination techniques to assess their usefulness. The methods are general and can be applied to any model‐derived predictand. A set of weather‐sensitive European country‐aggregate energy variables (electricity demand and wind power) is selected for demonstration purposes. Results show that these innovative methods exhibit significant skill improvements (i.e., between 5 and 15% improvement in the probabilistic skill) with respect to standard multimodel combination techniques for lead weeks up to 5. The incorporation of statistically derived predictors (based on historical climate data) alongside NWP forecasts is also shown to contribute significant skill improvements in many cases.