Bidirectional Recurrent Neural Network Research Articles

Ultra-deep channel sand body target recognition method based on improved deep learning under UAV cluster

Abstract River sand bodies have complex and changeable characteristics and distribution. In order to improve the accuracy and efficiency of target recognition, this study proposes a target recognition method of ultra-deep river sand bodies with improved deep learning under unmanned aerial vehicle (UAV) cluster. By constructing the cooperative target allocation model of UAV group, it is ensured that the targets of ultra-deep and large-area river sand bodies are collected. The gradient histogram is used to extract the image characteristics of ultra-deep river sand body and enhance the target image of ultra-deep river sand body. Bi-directional long short-term memory (Bi-LSTM) network model is constructed by introducing bidirectional recurrent neural network (RNN) to improve deep learning. Bi-LSTM neural network is used to construct the target recognition model of ultra-deep river sand body and complete the target recognition. The experimental results show that this method can extract the target edge completely and recognize the image edge accurately, and the average recognition accuracy under different ambiguities is higher than 95. It is proved that this method has high accuracy in sand body feature extraction and classification and has great application potential in river sand body target recognition.

Eye-LRCN: A Long-Term Recurrent Convolutional Network for Eye Blink Completeness Detection.

Computer vision syndrome causes vision problems and discomfort mainly due to dry eye. Several studies show that dry eye in computer users is caused by a reduction in the blink rate and an increase in the prevalence of incomplete blinks. In this context, this article introduces Eye-LRCN, a new eye blink detection method that also evaluates the completeness of the blink. The method is based on a long-term recurrent convolutional network (LRCN), which combines a convolutional neural network (CNN) for feature extraction with a bidirectional recurrent neural network that performs sequence learning and classifies the blinks. A Siamese architecture is used during CNN training to overcome the high-class imbalance present in blink detection and the limited amount of data available to train blink detection models. The method was evaluated on three different tasks: blink detection, blink completeness detection, and eye state detection. We report superior performance to the state-of-the-art methods in blink detection and blink completeness detection, and remarkable results in eye state detection.

Developing a hybrid system for stock selection and portfolio optimization with many-objective optimization based on deep learning and improved NSGA-III

Portfolio management is a critical aspect of investment strategies, with the goal to balance the low-risk and high-return investments. Despite this, existing portfolios frequently overlook the integration of stock selection outcomes and underutilize data from listed companies, leading to suboptimal portfolio performance. Addressing these shortcomings, this paper introduces a hybrid system involving stock selection and portfolio optimization. In stock selection, the system employs a combination of convolutional neural network and bi-directional recurrent neural network to predict stock trends. This approach enables the identification of stocks likely to appreciate in value, setting the stage for their inclusion in the subsequent optimization process. For portfolio optimization, the study formulates a five-objective optimization problem that incorporates mean, variance, skewness, kurtosis, and distance-to-default as key considerations. To solve the many-objective constrained optimization problem, an advanced strategy employing a static penalty function and an improved Non-dominated Sorting Genetic Algorithm III (NSGA-III) based on tent chaotic mapping is utilized. The efficacy of the proposed hybrid system is rigorously tested through three sets of ablation experiments alongside two discussions focused on its robustness and computational efficiency. The findings from these investigations reveal that the hybrid system outperforms traditional approaches, reducing risks and improving returns for investors.

Automatic Translation between Mixtec to Spanish Languages Using Neural Networks

This paper introduces a novel method for collecting and translating texts from the Mixtec to the Spanish language. The method comprises four primary steps. First, we collected a Mixtec–Spanish corpus that includes 4568 sentences from educational and religious domain texts. To enhance the parallel corpus, we generate synthetic data with GPT-3.5. Second, we cleaned the data with a semi-automatic approach followed by preprocessing and tokenization. In preprocessing, we removed stop words, duplicated sentences, special characters, and numbers and converted them to lowercase. Third, we performed semi-automatic alignment to find the correspondence of Mixtec–Spanish sentences to generate sentence-level aligned texts necessary for translation. Finally, we trained automatic translation models based on recurrent neural networks, bidirectional recurrent neural networks, and Transformers. Our system achieved a BLEU score of 95.66 for Mixtec-to-Spanish translation and 99.87 for Spanish-to-Mixtec translation. We also obtained a translation edit rate (TER) of 0.5 for Spanish-to-Mixtec and a TER of 16.5 for Mixtec-to-Spanish. Our research stands out as a pioneering effort in the field of automatic Mixtec-to-Spanish translation in Mexico, filling a gap identified in the current literature.

Triple Generative Self-Supervised Learning Method for Molecular Property Prediction.

Molecular property prediction is an important task in drug discovery, and with help of self-supervised learning methods, the performance of molecular property prediction could be improved by utilizing large-scale unlabeled dataset. In this paper, we propose a triple generative self-supervised learning method for molecular property prediction, called TGSS. Three encoders including a bi-directional long short-term memory recurrent neural network (BiLSTM), a Transformer, and a graph attention network (GAT) are used in pre-training the model using molecular sequence and graph structure data to extract molecular features. The variational auto encoder (VAE) is used for reconstructing features from the three models. In the downstream task, in order to balance the information between different molecular features, a feature fusion module is added to assign different weights to each feature. In addition, to improve the interpretability of the model, atomic similarity heat maps were introduced to demonstrate the effectiveness and rationality of molecular feature extraction. We demonstrate the accuracy of the proposed method on chemical and biological benchmark datasets by comparative experiments.

A general energy modeling network for serial industrial robots integrating physical mechanism priors

Industrial robots (IRs), as the core equipment of intelligent manufacturing, play increasingly important roles in various industrial scenarios such as assembly, welding, handling, and spraying, significantly improving production efficiency and product quality. The massive popularization and application of IRs have brought about a sharp increase in energy consumption (EC), and the modeling and optimization of EC is becoming imperative. In this paper, a general energy modeling network DM-PLM for serial IRs based on Dynamic Model (DM) and Power Loss Model (PLM) is proposed by integrating the prior knowledge of IR power composition and dynamic mechanism, enabling efficient and accurate modeling of dynamics, power, and EC under multi-load conditions. Considering the force transmission characteristics of serial robots, this paper proposes an improved bidirectional recurrent neural network (BiRNN) to model the joint dynamics. Additionally, a power loss model based on the ResNet convolutional neural network is employed. Experiments are carried out with a KUKA KR210 heavy-duty robot and a UR5 collaborative robot. The results show that the DM-PLM model incorporating the physical mechanism priors achieves 97 %, 98 %, and 99 % modeling accuracy in joint torques, total power, and EC for both robots under multi-load conditions. In addition, the proposed DM-PLM model is applied to the EC optimization of KUKA KR210 through trajectory planning, which achieves over 30 % EC reduction with the genetic algorithm, providing an effective approach to improving the energy efficiency of serial IRs.

CRISPR-M: Predicting sgRNA off-target effect using a multi-view deep learning network.

Using the CRISPR-Cas9 system to perform base substitutions at the target site is a typical technique for genome editing with the potential for applications in gene therapy and agricultural productivity. When the CRISPR-Cas9 system uses guide RNA to direct the Cas9 endonuclease to the target site, it may misdirect it to a potential off-target site, resulting in an unintended genome editing. Although several computational methods have been proposed to predict off-target effects, there is still room for improvement in the off-target effect prediction capability. In this paper, we present an effective approach called CRISPR-M with a new encoding scheme and a novel multi-view deep learning model to predict the sgRNA off-target effects for target sites containing indels and mismatches. CRISPR-M takes advantage of convolutional neural networks and bidirectional long short-term memory recurrent neural networks to construct a three-branch network towards multi-views. Compared with existing methods, CRISPR-M demonstrates significant performance advantages running on real-world datasets. Furthermore, experimental analysis of CRISPR-M under multiple metrics reveals its capability to extract features and validates its superiority on sgRNA off-target effect predictions.

Task sub-type states decoding via group deep bidirectional recurrent neural network

Decoding brain states under different cognitive tasks from functional magnetic resonance imaging (fMRI) data has attracted great attention in the neuroimaging filed. However, the well-known temporal dependency in fMRI sequences has not been fully exploited in existing studies, due to the limited temporal-modeling capacity of the backbone machine learning algorithms and rigid training sample organization strategies upon which the brain decoding methods are built. To address these limitations, we propose a novel method for fine-grain brain state decoding, namely, group deep bidirectional recurrent neural network (Group-DBRNN) model. We first propose a training sample organization strategy that consists of a group-task sample generation module and a multiple-scale random fragment strategy (MRFS) module to collect training samples that contain rich task-relevant brain activity contrast (i.e., the comparison of neural activity patterns between different tasks) and maintain the temporal dependency. We then develop a novel decoding model by replacing the unidirectional RNNs that are widely used in existing brain state decoding studies with bidirectional stacked RNNs to better capture the temporal dependency, and by introducing a multi-task interaction layer (MTIL) module to effectively model the task-relevant brain activity contrast. Our experimental results on the Human Connectome Project task fMRI dataset (7 tasks consisting of 23 task sub-type states) show that the proposed model achieves an average decoding accuracy of 94.7% over the 23 fine-grain sub-type states. Meanwhile, our extensive interpretations of the intermediate features learned in the proposed model via visualizations and quantitative assessments of their discriminability and inter-subject alignment evidence that the proposed model can effectively capture the temporal dependency and task-relevant contrast.

A deep learning-based imputation method for missing gaps in satellite aerosol products by fusing numerical model data

Satellite-based aerosol optical depth (AOD) products are commonly used in various aerosol-related studies, such as aerosol pollution mapping and aerosol-climate interactions. However, these satellite AOD products often suffer from significant missing gaps due to cloud cover and limitations in the retrieval algorithm. To address this issue, some studies take advantage of real-time seamless simulation of numerical models and successfully fill in these gaps by establishing a regression relationship between satellite AOD and numerical model AOD. However, these previous studies usually use satellite AOD retrievals as the regression target, which limits the accuracy of the imputation results by the original accuracy of satellite AOD retrievals and also consumes a considerable amount of time. To overcome these limitations, this study proposes a spatiotemporal imputation model called Bi-ConvRNN, which combines convolutional neural networks (CNN) and bidirectional recurrent neural networks (Bi-RNN). The model takes both satellite AOD retrievals and numerical model AOD data as input and utilizes the weighted mean squared error (MSE) loss function of multiple AOD datasets, e.g., ground-based data, satellite retrievals, and numerical simulation, as the optimization target to improve the imputation accuracy. The proposed model is evaluated using hourly COMS GOCI AOD products. In the independent test set, the AOD results generated by the Bi-ConvRNN model in the region containing GOCI AOD retrievals can break the accuracy of original GOCI AOD products with the accuracy improved from R2 = 0.70 [RMSE = 0.15] to R2 = 0.84 [RMSE = 0.11], and the filling accuracy, e.g. R2 = 0.79, [RMSE = 0.14], in the region without GOCI AOD retrievals are still better than those of the original GOCI AOD retrievals. Additionally, the Bi-ConvRNN model demonstrates satisfactory filling efficiency, requiring only 0.12 s to fill in the missing gaps of hourly GOCI AOD products per day. These results highlight the efficiency and reliability of the proposed model in filling the gaps in satellite AOD products, and the filled AOD results have great potential for further aerosol-related research.

Improved BTM topic embedding method for Web text data extraction

With the popularity of social software such as Facebook and Twitter, online text data has begun to explode. However, due to the sparseness and imbalance of text data, it is difficult to accurately analyze and process data such as emotions contained in online text. Therefore, according to the characteristics of Web text, the Web text data extraction model based on biterm topic model, bidirectional recurrent neural network, topic embedding and attention mechanism, namely BTM-BRNN-TEAM model, is proposed to improve the accuracy of emotion analysis in short text. The test results show that when the first 10 topic words are selected, the recognition accuracy, error rate, precision, recall rate and F1-measure of this model are about 77.6%, 22.4%, 83.4%, 76.4%, and 0.82, respectively, which are higher than those of the other two models. The area under the receiver operating characteristic curve of the model is about 0.91, which indicates that its performance is superior. In the processing of short text sentences, the proposed model is able to accurately identify aspectual words that have a significant impact on the text, while weakening the impact of non-aspectual words. The above results show that the Web text processing model proposed in this study has superior performance in sentiment analysis and semantic recognition.

Deep learning based assessment of hemodynamics in the coarctation of the aorta: comparison of bidirectional recurrent and convolutional neural networks.

The utilization of numerical methods, such as computational fluid dynamics (CFD), has been widely established for modeling patient-specific hemodynamics based on medical imaging data. Hemodynamics assessment plays a crucial role in treatment decisions for the coarctation of the aorta (CoA), a congenital heart disease, with the pressure drop (PD) being a crucial biomarker for CoA treatment decisions. However, implementing CFD methods in the clinical environment remains challenging due to their computational cost and the requirement for expert knowledge. This study proposes a deep learning approach to mitigate the computational need and produce fast results. Building upon a previous proof-of-concept study, we compared the effects of two different artificial neural network (ANN) architectures trained on data with different dimensionalities, both capable of predicting hemodynamic parameters in CoA patients: a one-dimensional bidirectional recurrent neural network (1D BRNN) and a three-dimensional convolutional neural network (3D CNN). The performance was evaluated by median point-wise root mean square error (RMSE) for pressures along the centerline in 18 test cases, which were not included in a training cohort. We found that the 3D CNN (median RMSE of 3.23mmHg) outperforms the 1D BRNN (median RMSE of 4.25mmHg). In contrast, the 1D BRNN is more precise in PD prediction, with a lower standard deviation of the error (±7.03mmHg) compared to the 3D CNN (±8.91mmHg). The differences between both ANNs are not statistically significant, suggesting that compressing the 3D aorta hemodynamics into a 1D centerline representation does not result in the loss of valuable information when training ANN models. Additionally, we evaluated the utility of the synthetic geometries of the aortas with CoA generated by using a statistical shape model (SSM), as well as the impact of aortic arch geometry (gothic arch shape) on the model's training. The results show that incorporating a synthetic cohort obtained through the SSM of the clinical cohort does not significantly increase the model's accuracy, indicating that the synthetic cohort generation might be oversimplified. Furthermore, our study reveals that selecting training cases based on aortic arch shape (gothic versus non-gothic) does not improve ANN performance for test cases sharing the same shape.

Image Identification and Error Correction Method for Test Report Based on Deep Reinforcement Learning and IoT Platform in Smart Laboratory

In order to solve the problems that most models are complex, time-consuming, and have difficulty in identifying image errors, an image identification and error correction method of test report based on deep reinforcement learning and the internet of things platform in the smart lab was proposed. Firstly, a smart lab architecture was designed based on the internet of things platform, achieving efficient operation of the laboratory through cloud edge collaboration. Then, the depth separable convolution improved convolutional neural network is used to extract image features, and the features are input into bidirectional recurrent neural networks (BiLSTM) for analysis to complete image recognition. Finally, the ICNN-BiLSTM model is used as the agent of reinforcement learning, and image error correction is completed by identifying the distance between the image and the key points of the reference image. Based on the Python platform, the proposed method was experimentally demonstrated, and the results showed that its average error correction accuracy reached 96.75%, with a processing time of 15.37s.

An reconstruction bidirectional recurrent neural network ‐based deinterleaving method for known radar signals in open‐set scenarios

AbstractIn electronic warfare, radar signal deinterleaving is a critical task. While many researchers have applied deep learning and utilised known radar classes to construct interleaved pulse sequences training sets for deinterleaving models, these models face challenges in distinguishing between known and unknown radar classes in open‐set scenarios. To address this challenge, the authors propose a novel model, the Reconstruction Bidirectional Recurrent Neural Network (RBi‐RNN). RBi‐RNN utilises input reconstruction and employs a joint training strategy incorporating cross‐entropy loss, reconstruction loss, and centre loss. These strategies aim to maximise inter‐class latent representation distances while minimising intra‐class disparities. By incorporating an open‐set recognition method based on extreme value theory, RBi‐RNN adapts to open‐set scenarios. Simulation results demonstrate the superiority of RBi‐RNN over conventional models in both closed‐set and open‐set scenarios. In open‐set scenarios, it successfully discriminates between known and unknown radar signals within interleaved pulse sequences, deinterleaving known radar classes with high stability. The authors lay the foundation for future unsupervised deinterleaving methods designed specifically for unknown radar pulses.

Short-term electricity load forecasting based on CEEMDAN-FE-BiGRU-Attention model

Abstract Intending to address the volatility and intricacy of power load data, a novel method for short-term power load forecasting is presented, utilizing the CEEMDAN-FE-BiGRU-Attention model. This paper introduces the innovative adaptive noise fully integrated Empirical Mode decomposition (CEEMDAN) algorithm, which effectively decomposes the sequence into modal components of various frequencies and residual components. This decomposition serves to reduce the complexity inherent in the original time series. Subsequently, the fuzzy entropy (FE) algorithm is employed to calculate the time complexity of each component, allowing for the reconstruction of different scale sequences, thereby enhancing computational efficiency. Finally, the sequences of varying scales are input into the bidirectional recurrent neural network (BiGRU) model, incorporating the attention mechanism for prediction. Notably, the prediction accuracy of the BiGRU model with attention mechanism surpasses that of the baseline GRU model for multifeature time series, resulting in a significant improvement in prediction accuracy. Experimental findings demonstrate that the proposed model outperforms traditional approaches, enabling better capture of the variation trends in power load data while reducing time series complexity. Moreover, the proposed model exhibits a remarkable reduction in mean absolute percentage error and root mean square error values by 90.24% and 85.05%, respectively, when compared to the single BiGRU-Attention model. This enhancement enhances the accuracy of power load prediction. These innovative methodologies endow the power load forecasting method proposed in this paper with greater potential and viability for real-world applications.

Multichannel Nonlinear Equalization in Coherent WDM Systems Based on Bi-Directional Recurrent Neural Networks

Multichannel Nonlinear Equalization in Coherent WDM Systems Based on Bi-Directional Recurrent Neural Networks

Optimizing RNNs for EMG Signal Classification: A Novel Strategy Using Grey Wolf Optimization.

Accurate classification of electromyographic (EMG) signals is vital in biomedical applications. This study evaluates different architectures of recurrent neural networks for the classification of EMG signals associated with five movements of the right upper extremity. A Butterworth filter was implemented for signal preprocessing, followed by segmentation into 250 ms windows, with an overlap of 190 ms. The resulting dataset was divided into training, validation, and testing subsets. The Grey Wolf Optimization algorithm was applied to the gated recurrent unit (GRU), long short-term memory (LSTM) architectures, and bidirectional recurrent neural networks. In parallel, a performance comparison with support vector machines (SVMs) was performed. The results obtained in the first experimental phase revealed that all the RNN networks evaluated reached a 100% accuracy, standing above the 93% achieved by the SVM. Regarding classification speed, LSTM ranked as the fastest architecture, recording a time of 0.12 ms, followed by GRU with 0.134 ms. Bidirectional recurrent neural networks showed a response time of 0.2 ms, while SVM had the longest time at 2.7 ms. In the second experimental phase, a slight decrease in the accuracy of the RNN models was observed, standing at 98.46% for LSTM, 96.38% for GRU, and 97.63% for the bidirectional network. The findings of this study highlight the effectiveness and speed of recurrent neural networks in the EMG signal classification task.

Short-term drought Index forecasting for hot and semi-humid climate Regions: A novel empirical Fourier decomposition-based ensemble Deep-Random vector functional link strategy

The development of advanced technologies based on computer aid models in the domain of crops and agriculture productively is a modern advancement. Machine learning (ML) based forecasting of the short-term drought indicators (e.g., Standardized Precipitation Evapotranspiration Index (SPEI)) based on individual signals is a complex process that involves several factors including data quality and availability, inherent signal complexity, non-stationarity of climate, and uncertainty in ML models. Recent achievements in the field of Fourier-based signal processing integrated with advanced deep learning approaches have made it possible to produce very accurate intelligent frameworks for multi-temporal drought indicators and fill this gap. In this research, a new ultramodern complementary intelligent framework comprised of the SelectKbest feature selection (FS), Empirical Fourier Decomposition (EFD), and deep ensemble random vector functional link (Deep RVFL) was developed for multi-temporal monthly forecasting of short-term drought indicators for three and six months (SPEI3 and SPEI6) for two different very hot and semi-humid climate zones of Iran.. For this purpose, the most influential time lags associated with each drought indicator were indicated using the SelectKbest FS in each zone. Afterwards, the individual SPEI signals were decomposed by the EFD technique imposing the most important lagged components to feed the ML approaches. Here, a new hybrid architecture deep learning model, namely a convolutional neural network coupled with a bidirectional gated recurrent unit (CNN-Bi-GRU) designed for multi-temporal drought forecasting in the next one-and three- months. Two advanced approaches, introducing the convolutional neural network coupled with bidirectional recurrent neural network (CNN-Bi-RNN), and Random vector function link (RVFL) were adopted to validate the main model (EFD-DeepRVFL) in complementary and standalone counterpart forms. The robustness of all the models was examined using several metrics such as coefficient of determination (R2), root mean square error (RMSE), reliability, and squared Chi-square Distance (SquD). The comprehensive assessment of the outcomes of hybrid schemes revealed that EFD-DeepRVFL owing to superior performance (R2|SPEI3(t + 1) = 0.953, R2|SPEI3(t + 3) = 0.837, R2|SPEI6(t + 1) = 0.962, and R2|SPEI6(t + 3) = 0.887) at Ahvaz station and (R2|SPEI3(t + 1) = 0.964, R2|SPEI3(t + 3) = 0.863, R2|SPEI6(t + 1) = 0.935, and R2|SPEI6(t + 3) = 0.839) at Kermanshah station outperformed the EFD-RVFL and CNN-Bi-RNN, respectively. The developed expert system provides early warning of drought conditions, as a decision-making tool, crop yield prediction, and water resources risk assessment.

Denoising sparker seismic data with Deep BiLSTM in fractional Fourier transform

Sparkers are widely-used sources of high-resolution images for shallow sub-bottom seismic monitoring. Sparker seismic data contain high random noise levels caused by the high voltage and electrical current required to power the sparker source. Random noise is distributed over the entire frequency and wavenumber, and is thereby difficult to suppress. Numerous studies have been conducted to develop random noise suppression by using convolutional neural networks (CNNs). However, because CNNs learn common characteristics of random noise without considering the sequence while training, learning the sequence relationship between useful signal and random noise in time-series data is difficult. As an alternative, we proposed a recurrent neural network (RNN)-based denoising method that considers seismic data sequences. Effective denoising of random noise using RNNs requires features that distinguish between useful signals and random noise during network training. We proposed a denoising method for sparker seismic data that uses bidirectional long short-term memory RNN in fractional Fourier transform (FrFT). The time-frequency distribution in the FrFT domain was utilized for network training, and useful signals were effectively separated from random noise. The proposed method was applied to multi-channel sparker seismic data acquired in Yeongil, Pohang (offshore eastern Korea) by the Korea Maritime & Ocean University. Numerical results confirmed that the proposed method effectively suppressed random noise while maintaining a useful signal in sparker seismic data.

Towards a Bidirectional Mexican Sign Language–Spanish Translation System: A Deep Learning Approach

People with hearing disabilities often face communication barriers when interacting with hearing individuals. To address this issue, this paper proposes a bidirectional Sign Language Translation System that aims to bridge the communication gap. Deep learning models such as recurrent neural networks (RNN), bidirectional RNN (BRNN), LSTM, GRU, and Transformers are compared to find the most accurate model for sign language recognition and translation. Keypoint detection using MediaPipe is employed to track and understand sign language gestures. The system features a user-friendly graphical interface with modes for translating between Mexican Sign Language (MSL) and Spanish in both directions. Users can input signs or text and obtain corresponding translations. Performance evaluation demonstrates high accuracy, with the BRNN model achieving 98.8% accuracy. The research emphasizes the importance of hand features in sign language recognition. Future developments could focus on enhancing accessibility and expanding the system to support other sign languages. This Sign Language Translation System offers a promising solution to improve communication accessibility and foster inclusivity for individuals with hearing disabilities.

Quantitative improvement of streamflow forecasting accuracy in the Atlantic zones of Canada based on hydro-meteorological signals: A multi-level advanced intelligent expert framework

Developing reliable streamflow forecasting models is critical for hydrological tasks such as improving water resource management, analyzing river patterns, and flood forecasting. In this research, for the first time, an emerging multi-level TOPSIS (technique for order preference by similarity to the ideal solution) -based hybridization comprised of the Boruta classification and regression tree (Boruta-CART) feature selection, multivariate variational mode decomposition (MVMD), and a hybrid Convolutional Neural Network (CNN) Bidirectional Gated Recurrent Unit (CNN-BiGRU) deep learning was adopted to multi-temporal (one and three days ahead) forecast the daily streamflow in the Rivers of Prince Edward Island, Canada. For this aim, in the first step, the Boruta-CART feature selection technique determines the most effective lagged components among all the antecedent two-day information (i.e., t-1 and t-2) of hydro-meteorological features (from 2015 to 2020), including the water level, mean air temperature, heat degree days, total precipitation, dew point temperature, and relative humidity in the Bear and Winter Rivers of Prince Edward Island, Canada. Afterwards, a multivariate variational mode decomposition (MVMD) decomposes the input time series to decrease the complexity and non-linearity of the non-stationary ones before feeding the deep learning (DL) models. Here, the CNN-GRU was employed as the primary DL model, along with the kernel extreme machine method (KELM), random variational function link (RVFL), and hybrid CNN bidirectional recurrent neural network (CNN-BiRNN) as the comparative models. A TOPSIS scheme applying several performance measures like the correlation coefficient (R), root mean square error (RMSE), and reliability was designed for the robustness assessment of the hybrid (MVM-CNN-BiGRU, MVM-CNN-BiRNN, MVM-RVFL, and MVM-KELM) and standalone models. The computational outcomes revealed that in the Bear River, the MVM-CNN-BiGRU, owing to its best forecasting performance (one day ahead: TOPSIS score 1, R = 0.960, RMSE = 0.098, and reliability = 65.082; three days ahead: TOPSIS score = 0.999, R = 0.924, and RMSE = 0.33) outperformed the other hybrid models, followed by the MVM-CNN-BiRNN, MVM-RVFL, and MVM-KELM, respectively. Moreover, in the Winter River, the MVM-CNN-BiGRU in terms of (one-day ahead: TOPSIS score = 0.890, R = 0.955, RMSE = 0.274, and reliability = 34.004; three-days ahead: TOPSIS score = 0.686, R = 0.924, and RMSE = 0.330) was superior to the other models. The provided expert system could be vital in the local flood decision-making process, in the absence of streamflow information as input modeling, during the flood seasons to reduce flood damage in residential areas.