Input Dataset Research Articles

Концепция использования искусственного интеллекта в автоматизированных системах управления твердыми бытовыми отходами

In the Republic of Kazakhstan, there is currently no proper control and accounting of solid waste by the state and culture, a high standard of living for the majority of the population. To solve the problem of automated MSW management using AI, it is proposed to develop a knowledge base, the level of education, distribution by collection zones in the city, socio-economic stratification, population and the amount of solid waste generated over a certain period of time is taken as a set of input data. The initial filling of the knowledge base is provided according to the latest population census of the Republic of Kazakhstan https://stat. gov.kz/ru/national/2021/. For predictive estimates of the volume of solid waste, the method of the forest conveyor with Bayesian optimization (RFBO) was used, the algorithm and architecture of the software solution are quite informative. The reliability of predictive decisions on the volume of solid waste presented by the program is at least 90%, taking into account the correctness of the input data set, the program code is written in Python. The solution proposed in the article to the problem related to the automation of solid waste management is based on international experience, taking into account the identified shortcomings, as well as the prospects and trends of understanding by residents of the country and public utilities of the country, a possible garbage collapse. The possible disadvantages of the program are the availability of constant access to the Internet, the need for a large amount of RAM, and computer performance. The program can be used as a ready-made solution for predictive estimates of solid waste for various regions

Deep hybrid model for attack detection in IoT-fog architecture with improved feature set and optimal training

IoT-Fog computing provides a wide range of services for end-based IoT systems. End IoT devices interface with cloud nodes and fog nodes to manage client tasks. Critical attacks like DDoS and other security risks are more likely to compromise IoT end devices while they are collecting data between the fog and the cloud layer. It’s important to find these network vulnerabilities early. By extracting features and placing the danger in the network, DL is crucial in predicting end-user behavior. However, deep learning cannot be carried out on Internet of Things devices because to their constrained calculation and storage capabilities. In this research, we suggest a three-stage Deep Hybrid Detection Model for Attack Detection in IoT-Fog Architecture. Improved Z-score normalization-based data preparation will be carried out in the initial step. On the basis of preprocessed data, features like IG, raw data, entropy, and enhanced MI are extracted in the second step. The collected characteristics are used as input to hybrid classifiers dubbed optimized Deep Maxout and Deep Belief Network (DBN) in the third step of the process to classify the assaults based on the input dataset. A hybrid optimization model called the BMUJFO (Blue Monkey Updated Jellyfish Optimization) technique is presented for the best Deep Maxout training. Additionally, the suggested model produced higher accuracy, precision, sensitivity, and specificity results, with values of 95.26 percent, 94.84%, 96.28%, and 97.84%, respectively.

Transfer Learning with ResNet3D-101 for Global Prediction of High Aerosol Concentrations

In order to better predict the high aerosol concentrations associated with air pollution and climate change, a machine learning model was developed using transfer learning and the segmentation process of global satellite images. The main concept of transfer learning lies on convolutional neural networks and works by initializing the already trained model weights to better adapt the weights when the network is trained on a different dataset. The transfer learning technique was tested with the ResNet3D-101 model pre-trained from a 2D ImageNet dataset. This model has performed well for contrail detection to assess climate impact. Aerosol distributions can be monitored via satellite remote sensing. Satellites can monitor some aerosol optical properties like aerosol optical thickness. Aerosol optical thickness snapshots were the input dataset for the model and were obtained from NASA’s Terra-Modis satellite; the output images were segmented by comparing the pixel values with a threshold value of 0.8 for aerosol optical thickness. Hyperparameter optimization finds a tuple of hyperparameters that yields an optimal model that minimizes a predefined loss function on given independent data. The model structure was adjusted in order to improve the performance of the model by applying methods and hyperparameter optimization techniques such as grid search, batch size, threshold, and input length. According to the criteria defined by the authors, the distance domain criterion and time domain criterion, the developed model is capable of generating adequate data and finding patterns in the time domain. As observed from the comparison of relative coefficients for the criteria metrics proposed by the authors, ddc and dtc, the deep learning model based on ConvLSTM layers developed in our previous studies has better performance than the model developed in this study with transfer learning.

Experimental analysis of forming forces in incremental forming and its prediction by AI-based methods

Single point incremental forming (SPIF) is an emerging approach to manufacture the user-specific parts of greater strength economically without involving dedicated tools and dies. In this study, peak deforming loads have been measured and estimated in order to determine the size and capacity of forming machinery for manufacturing components, using the SPIF process, for designed process conditions on AA7075-O sheets. Estimation of maximum deforming loads has been accomplished with the help of hybrid artificial neural network (HANN) approach by taking experimental load values as the input dataset. The load values, estimated by HANN approach, have been compared with the load values estimated by two state-of-art regression approaches, that is, Support Vector Machine (SVM) and Gaussian Process Regression (GPR). Results show that the regression methods (SVM and GPR) are outperformed by the HANN model with an accuracy of 99.80%. HANN model estimated force values very close to experimental results. The forming forces were found to be greater when the flat-end tools were used as a forming agent in place of hemispherical-end tool. The proposed study would help the researchers and scientists to assess the capacity the machines used in SPIF process to prepare a particular sheet-part and to prevent the failure of hardware during the SPIF process.

Multilayer adaptive critic design with digital twin for data-driven optimal tracking control and industrial applications

In this paper, an optimal trajectory tracking control problem for general nonlinear systems is investigated. An adaptive critic control method with the digital twin (DT) theory is developed. Divergent from the existing tracking control methods, the advantages of adaptive dynamic programming (ADP) and the theory of DT are combined in this paper, and the novel multilayer artificial system structure is constructed. The action-critic structure is employed by each artificial system to obtain an approximate optimal control policy. The model network (MN) is built by using the actual input and output data sets of the controlled system, which means the dependence on the dynamics of the system is overcame. Then, the weights of the trained action network (AN) and MN are passed to the real system to realize the optimal tracking control. The feasibility of the algorithm is proved by theoretical analysis. Finally, the algorithm is applied to a simple nonlinear torsional pendulum system and an industrial wastewater treatment system (WWTS), and the effectiveness of the algorithm is verified. The algorithm effectively realizes the tracking control of nonlinear systems.

Changes in External Forcings Drive Divergent AMOC Responses Across CESM Generations

AbstractIt has been suggested that the Atlantic meridional overturning circulation (AMOC) in many CMIP6 models is overly sensitive to anthropogenic aerosol forcing, and it has been proposed that this is due to the inclusion of aerosol indirect effects for the first time in many CMIP6 models. We analyze the AMOC response in a newly released ensemble of simulations performed with CESM2 forced by the CMIP5 input data sets (CESM2‐CMIP5). This AMOC response is then compared to the CMIP5‐generation CESM1 large ensemble (CESM1‐LE) and the CMIP6‐generation CESM2 large ensemble (CESM2‐LE). A key conclusion, only made possible by this experimental setup, is that changes in aerosol‐indirect effects cannot explain differences in AMOC response between CESM1‐LE and CESM2‐LE. Instead, we hypothesize that the difference is due to increased interannual variability of anthropogenic emissions. This forcing variability may act through a nonlinear relationship between the surface heat budget of the North Atlantic and the AMOC.

Gold price prediction by a CNN-Bi-LSTM model along with automatic parameter tuning.

Banking and stock markets consider gold to be an important component of their economic and financial status. There are various factors that influence the gold price trend and its fluctuations. Accurate and reliable prediction of the gold price is an essential part of financial and portfolio management. Moreover, it could provide insights about potential buy and sell points in order to prevent financial damages and reduce the risk of investment. In this paper, different architectures of deep neural network (DNN) have been proposed based on long short-term memory (LSTM) and convolutional-based neural networks (CNN) as a hybrid model, along with automatic parameter tuning to increase the accuracy, coefficient of determination, of the forecasting results. An illustrative dataset from the closing gold prices for 44 years, from 1978 to 2021, is provided to demonstrate the effectiveness and feasibility of this method. The grid search technique finds the optimal set of DNNs' parameters. Furthermore, to assess the efficiency of DNN models, three statistical indices of RMSE, RMAE, and coefficient of determination (R2), were calculated for the test set. Results indicate that the proposed hybrid model (CNN-Bi-LSTM) outperforms other models in total bias, capturing extreme values and obtaining promising results. In this model, CNN is used to extract features of input dataset. Furthermore, Bi-LSTM uses CNN's outputs to predict the daily closing gold price.

AutoML‐based predictive framework for predictive analysis in adsorption cooling and desalination systems

AbstractAdsorption cooling and desalination systems have a distinct advantage over other systems that use low‐grade waste heat near ambient temperature. Since improving their performance, including reliability and failure prediction, is challenging, developing an efficient diagnostic system is of great practical significance. The paper introduces artificial intelligence (AI) and an automated machine learning approach (AutoML) in a real‐life application for a computational diagnostic system of existing adsorption cooling and desalination facilities. A total of 1769 simulated data points containing data indicating a failure status are applied to develop a comprehensive AI‐based Diagnostic (AID) system covering a wide range of 42 input parameters. The paper introduces a conditional monitoring system for adsorption cooling and desalination systems. The novelty of the presented study mainly consists of two aspects. First, the intelligent system predicts the health or failure states of various components in a complex three‐bed adsorption chiller installation using the extensive input data sets of 42 different operating parameters. The developed AID expert tool, based on selecting the best from 42 models generated by the DataRobot platform, was validated on the complex, existing three‐bed adsorption chiller. The AID system correctly identified healthy and failure states in various installation components. The developed expert system is very efficient (AUC = 0.988, RMSE = 0.20, LogLoss = 0.14) in predicting emergency states. The proposed method constitutes a quick and easy technique for failure prediction and represents a complementary tool compared to the other condition monitoring methods.

Machine learning in the classification of asian rust severity in soybean using hyperspectral sensor

Traditional monitoring of asian soybean rust severity is a time- and labor-intensive task, as it requires visual assessments by skilled professionals in the field. Thus, the use of remote sensing and machine learning (ML) techniques in data processing has emerged as an approach that can increase efficiency in disease monitoring, enabling faster, more accurate and time- and labor-saving evaluations. The aims of the study were: (i) to identify the spectral signature of different levels of Asian soybean rust severity; (ii) to identify the most accurate machine learning algorithm for classifying disease severity levels; (iii) which spectral input provides the highest classification accuracy for the algorithms; (iv) to determine a sample size of leaves that guarantees the best accuracy for the algorithms. A field experiment was carried out in the 2022/2023 harvest in a randomized block design with a 6x3 factorial scheme (ML algorithms x severity levels) and four replications. Disease severity levels assessed were: healthy leaves, 25 % severity, and 50 % severity. Leaf hyperspectral analysis was carried out over a wide range from 350 to 2500 nm. From this analysis, 28 spectral bands were extracted, seeking to distinguish the spectral signature for each severity level with the least input dataset. Data was subjected to machine learning analysis using Artificial Neural Network (ANN), REPTree (DT) and J48 decision trees, Random Forest (RF), and Support Vector Machine (SVM) algorithms, as well as a traditional classification method (Logistic Regression - LR). Two different input datasets were tested for each algorithm: the full spectrum (ALL) provided by the sensor and the 28 spectral bands (SB). Tests with different sample sizes were also conducted to investigate the algorithms' ability to detect severity levels with a reduced sample size. Our findings indicate differences between the spectral curves for the severity levels assessed, which makes it possible to differentiate between healthy plants with low and high severity using hyperspectral sensing. SVM was the most accurate algorithm for classifying severity levels by using all the spectral information as input. This algorithm also provided high classification accuracy when using smaller leaf samples. This study reveals that hyperspectral sensing and the use of ML algorithms provide an accurate classification of different levels of Asian rust severity, and can be powerful tools for a more efficient disease monitoring process.

Intra-class deep learning object detection on embedded computer system

<span lang="EN-US">Implementation of artificial intelligence tends to be portable, mobile and embeds in embedded computer system (EBD). EBD is a special-purpose computer with limited capacity in a small-form size. Deep learning (DL) had known as cutting edges for object recognition. With DL, object feature extraction analysis is omitted. DL requires large computing resources and capacity. Implement DL algorithm on EBD goal to achieves high detection accuracy and high-efficiency resources. Hence, be able to cope with intra-class variations, and image disturbances. By those challenges and limitations, this study reports the performance of EBD to recognize an object which has high variations in their class, through an optimal raw-input dataset. The raw-input dataset performed optimization process with a supervisor. Yield is the proper optimal input dataset in size. The performance results observed begin from training dataset until evaluation stage of DL. The comparison performs in efficiency resources, loss, validation-loss, timesteps, and detection accuracy by multiclass confusion matrix analysis. This study shows through this purpose method efficient resources are highly archived. Shorter timesteps ensure training stage is successful, and detection accuracy is perfectly archived. In addition, this study proves DL method archived great performances in classifying object that has identical structure.</span>

Weather and Crop Yield Prediction by Machine Learning Model and Spiking Neural Network

This paper, "Weather and Crop Yield Prediction by Machine Learning Model and Spiking Neural Network," is to improve crop yield prediction efficiency by utilizing cutting-edge technology and transform agricultural practices. The creation of two prediction models specifically for this purpose is the main goal of the study. The first model incorporates geographical information such as Normalized difference vegetation index (NDVI), Standard Precipitation Index (SPI), and Vector Choquet Integral (VCI) and uses a Genetic Algorithm to reduce the size of the input dataset. This reduction attempts to lessen learning process confusion brought on by comparable data with different values. The model improves its capacity to identify pertinent patterns for precise yield estimates by incorporating spatial data. The SoftMax function is employed by the second model, a Modified Convolutional Neural Network (MCNN), as an error correcting method. By reducing errors and raising the overall accuracy of the model, the SoftMax function improves the output and produces forecasts that are more accurate. The study emphasizes how crucial geographic data including satellite imagery is for supplying crucial insights for predicting crop yields. Farmers can make well-informed decisions by having a thorough understanding of agricultural dynamics, which is facilitated by spatial patterns and temporal shifts. Statistical and geographic data are used to completely study the effects of numerous variables, such as biological, economic, and environmental ones, on crop yield production. The goal of the project is to create precise and effective instruments that enable farmers to make knowledgeable decisions about their soil by combining machine learning with spiking neural networks.

Artificial Neural Network-based robust technique for period prediction of Ottoman minarets in Türkiye

Minarets are crucial as symbolic and indispensable elements in mosques across religious communities worldwide. Beyond architectural culture, hazardous conditions have shaped the minaret structural formations in Islamic geography. Among these, masonry minarets built during the Ottoman Empire faced significant vulnerability to lateral forces due to their slender design, brittle materials, and distinctive shapes. After many earthquakes experienced in the 20th and 21st centuries, the collapse of numerous minarets highlights the urgent need to preserve the remaining structures in high seismicity regions of the historical Ottoman lands, moreover, identifying and fortifying the most vulnerable ones takes precedence. This study proposes an innovative approach, departing from previous research that solely focused on the correlation between geometrical and material properties such as height, cross-section area, moment of inertia, Young’s modulus, and material density using approximate formulas to predict the fundamental periods. Instead, it introduces a novel technique based on Artificial Neural Networks (ANNs) to predict the first three periods of Ottoman masonry minarets. Accurate measurements from 18 minarets located in Bursa City, Türkiye, were meticulously collected under ambient conditions, serving as the foundation for the neural network's comprehensive output database. The distinct parameters (geometrical and material properties) of these minarets form the input dataset. To mitigate the influence of random measurement noise, an effective averaging scheme was implemented. As a result, the proposed ANN technique demonstrates its robustness and holds great promise for practical applications, as it enables accurate estimation of the desired modal information for both the 18 minarets used to train the networks and the remaining three minarets, achieving high levels of accuracy.

Predictive and experimental analysis of forces in die-less forming using artificial intelligence techniques

The use of a novel technology for producing the components of lightweight materials and to reduce the requirements of power utilized during manufacturing processes can be a great aspect to decrease pollution and save resources. Single point incremental forming (SPIF) is the viable and novel approach for manufacturing the parts of high strength and lightweight materials without involving dedicated tools and dies economically. This die-less forming technique outperforms the conventional forming techniques by saving the energy and materials. In this work, the estimation and investigation of forming forces have been accomplished to ensure the secure uses for the SPIF machines for performing this process for the designed conditions on AA2024 sheets which is a lightweight aluminum alloy being widely used in aerospace and automotive sectors. To predict the peak deforming load, machine learning (ML) techniques are employed in the current work along with the artificial neural network (ANN) by taking experimental results as the input dataset. The proposed ML model revealed better accuracy (99%) than previous work performed using similar approaches. The proposed ANN model produced lower mean absolute percentage error 4.35 as compared to other models. Authors also calculated the computing time taken during estimation of forming force. Combination of the Flatend-R1 tool and the 1.6 mm blank thickness increased the deforming loads drastically and can become the limiting factor for forming machine which should be avoided whereas the combination of hemispherical tool and lower blank thickness (0.5 mm) reduced the deforming loads that are needed to manufacture the conical frustum. It was also noticed that as the tool shape was changed from hemispherical-end to Flatend-R1, the axial peak forces were increased by 13.16%, 16.59%, 20.43%, and 22.78% for the levels 1, 2, 3, and 4 of the blank thickness, respectively.

A Multiple Criteria Decision-Making Method Generated by the Space Colonization Algorithm for Automated Pruning Strategies of Trees

The rise of mechanical automation in orchards has sparked research interest in developing robots capable of autonomous tree pruning operations. To achieve accurate pruning outcomes, these robots require robust perception systems that can reconstruct three-dimensional tree characteristics and execute appropriate pruning strategies. Three-dimensional modeling plays a crucial role in enabling accurate pruning outcomes. This paper introduces a specialized tree modeling approach using the space colonization algorithm (SCA) tailored for pruning. The proposed method extends SCA to operate in three-dimensional space, generating comprehensive cherry tree models. The resulting models are exported as normalized point cloud data, serving as the input dataset. Multiple criteria decision analysis is utilized to guide pruning decisions, incorporating various factors such as tree species, tree life cycle stages, and pruning strategies during real-world implementation. The pruning task is transformed into a point cloud neural network segmentation task, identifying the trunks and branches to be pruned. This approach reduces the data acquisition time and labor costs during development. Meanwhile, pruning training in a virtual environment is an application of digital twin technology, which makes it possible to combine the meta-universe with the automated pruning of fruit trees. Experimental results demonstrate superior performance compared to other pruning systems. The overall accuracy is 85%, with mean accuracy and mean Intersection over Union (IoU) values of 0.83 and 0.75. Trunks and branches are successfully segmented with class accuracies of 0.89 and 0.81, respectively, and Intersection over Union (IoU) metrics of 0.79 and 0.72. Compared to using the open-source synthetic tree dataset, this dataset yields 80% of the overall accuracy under the same conditions, which is an improvement of 6%.

Automated personalization of biomechanical knee model.

Patient-specific biomechanical models of the knee joint can effectively aid in understanding the reasons for pathologies and improve diagnostic methods and treatment procedures. For deeper research of knee diseases, the development of biomechanical models with appropriate configurations is essential. In this study, we mainly focus on the development of a personalized biomechanical model for the investigation of knee joint pathologies related to patellar motion using automated methods. This study presents a biomechanical model created for patellar motion pathologies research and some techniques for automating the generation of the biomechanical model. To generate geometric models of bones, the U-Net neural network was adapted for 3D input datasets. The method uses the same neural network for segmentation of femur, tibia, patella and fibula. The total size of the train/validation (75/25%) dataset is 18,183 3D volumes of size voxels. The configuration of the biomechanical knee model proposed in the paper includes six degrees of freedom for the tibiofemoral and patellofemoral joints, lateral and medial contact surfaces for femur and tibia, and ligaments, representing, among other things, the medial and lateral stabilizers of the knee cap. The development of the personalized biomechanical model was carried out using the OpenSim software system. The automated model generation was implemented using OpenSim Python scripting commands. The neural network for bones segmentation achieves mean DICE 0.9838. A biomechanical model for realistic simulation of patellar movement within the trochlear groove was proposed. Generation of personalized biomechanical models was automated. In this paper, we have implemented a neural network for the segmentation of 3D CT scans of the knee joint to produce a biomechanical model for the study of knee cap motion pathologies. Most stages of the generation process have been automated and can be used to generate patient-specific models.

Event-Triggered Relearning Modeling Method for Stochastic System with Non-Stationary Variable Operating Conditions

This study presents a novel event-triggered relearning framework for neural network modeling, designed to improve prediction precision in dynamic stochastic complex industrial systems under non-stationary and variable conditions. Firstly, a sliding window algorithm combined with entropy is applied to divide the input and output datasets across different operational conditions, establishing clear data boundaries. Following this, the prediction errors derived from the neural network under different operational states are harnessed to define a set of event-triggered relearning criteria. Once these conditions are triggered, the relevant dataset is used to recalibrate the model to the specific operational condition and predict the data under this operating condition. When the predicted data fall within the training input range of a pre-trained model, we switch to that model for immediate prediction. Compared with the conventional BP neural network model and random vector functional-link network, the proposed model can produce a better estimation accuracy and reduce computation costs. Finally, the effectiveness of our proposed method is validated through numerical simulation tests using nonlinear Hammerstein models with Gaussian noise, reflecting complex stochastic industrial processes.

Development of single point prediction model using artificial neural network and experimental validation for pump as turbine applications

This paper proposed a new approach based on artificial neural networks (ANNs) to select the most appropriate centrifugal pumps for reverse operation i.e. pump as turbines (PAT), by predicting parameters at Best Efficiency Point from pump mode data. Both, exhaustive experimental data of 49 pumps (specific speeds 9−104) collected from open literature and by in-house experimentation data, considered input data-set as pump and target data-set as PAT are used for training the ANN models with two different training functions: Levenberg–Marquardt and Bayesian regularisation. The proposed ANN prediction model shows a deviation lower than 10% compared to the respective experimental value. Additionally, the trained ANN model tested with four pumps (not included in training data sets of proposed models) shows maximum absolute deviation in head number, flow number, and efficiency within the 10%, 7%, and 6% range, respectively, compared to the experimental values. Furthermore, the efficiency of reverse mode evaluated for variation in rotational speed shows a maximum of 3.3% (within 5%) absolute deviation compared to respective experimental results. Overall, the ANNs based prediction model of PAT parameters is recommended compared to conventional models available in the literature as it gives superior results (less than 10% deviation) for practical application.

IDaPS — Improved data-locality aware data placement strategy based on Markov clustering to enhance MapReduce performance on Hadoop

The execution of Map-Reduce applications on the Hadoop cluster poses significant challenges due to the non-consideration of data locality, i.e., assigning tasks to compute nodes where input data sets are located. Due to such non-consideration, high data transfer overheads are caused. Further, it increases latency, which may arise if input data needs to be transferred across the network, thereby significantly increasing execution time. To address this issue, an Improved DAta Placement Strategy IDaPS based on the intra-dependency among the data is proposed. IDaPS re-organizes the default data layouts in HDFS to ensure higher degree of parallelism. The efficiency of IDaPS is demonstrated in Hadoop clusters (10 and 15 nodes) by executing Hadoop Benchmark performance tests viz. WordCount, Grep on Project-Gutenberg book dataset (50 GB) and Least Square Linear Regression (LSLR) on weather dataset (10.67 GB). The results were compared with state-of-the-art algorithms viz. Hadoop Default Data Placement (HDDP), Load-Balancer and literary work RENDA. The results demonstrate that IDaPS significantly reduces execution time by 28.2% and 38.4% in 10-node and 15-node clusters while executing WordCount, and 35% and 38.1% in 10-node and 15-node clusters for Grep. Similarly, for LSLR, it reduces execution time by 32.7%.

A hybrid thyroid tumor type classification system using feature fusion, multilayer perceptron and bonobo optimization.

Thyroid tumor is considered to be a very rare form of cancer. But recent researches and surveys highlight the fact that it is becoming prevalent these days because of various factors. This paper proposes a novel hybrid classification system that is able to identify and classify the above said four different types of thyroid tumors using high end artificial intelligence techniques. The input data set is obtained from Digital Database of Thyroid Ultrasound Images through Kaggle repository and augmented for achieving a better classification performance using data warping mechanisms like flipping, rotation, cropping, scaling, and shifting. The input data after augmentation goes through preprocessing with the help of bilateral filter and is contrast enhanced using dynamic histogram equalization. The ultrasound images are then segmented using SegNet algorithm of convolutional neural network. The features needed for thyroid tumor classification are obtained from two different algorithms called CapsuleNet and EfficientNetB2 and both the features are fused together. This process of feature fusion is carried out to heighten the accuracy of classification. A Multilayer Perceptron Classifier is used for classification and Bonobo optimizer is employed for optimizing the results produced. The classification performance of the proposed model is weighted using metrics like accuracy, sensitivity, specificity, F1-score, and Matthew's correlation coefficient. It can be observed from the results that the proposed multilayer perceptron based thyroid tumor type classification system works in an efficient manner than the existing classifiers like CANFES, Spatial Fuzzy C means, Deep Belief Networks, Thynet and Generative adversarial network and Long Short-Term memory.

Deep Convolutional LSTM for improved flash flood prediction

Flooding remains one of the most devastating and costly natural disasters. As flooding events grow in frequency and intensity, it has become increasingly important to improve flood monitoring, prediction, and early warning systems. Recent efforts to improve flash flood forecasts using deep learning have shown promise, yet commonly-used techniques such as long short term memory (LSTM) models are unable to extract potentially significant spatial relationships among input datasets. Here we propose a hybrid approach using a Convolutional LSTM (ConvLSTM) network to predict stream stage heights using multi-modal hydrometeorological remote sensing and in-situ inputs. Results suggest the hybrid network can more effectively capture the specific spatiotemporal landscape dynamics of a flash flood-prone catchment relative to the current state-of-the-art, leading to a roughly 26% improvement in model error when predicting elevated stream conditions. Furthermore, the methodology shows promise for improving prediction accuracy and warning times for supporting local decision making.