Research on Time Series-Based Pipeline Ground Penetrating Radar Calibration Angle Prediction Algorithm.

The pipeline ground-penetrating radar stands as an indispensable detection device for ensuring underground space security. A wheeled pipeline robot is deployed to traverse the interior of urban underground drainage pipelines along their central axis. It is subject to influences such as resistance, speed, and human factors, leading to deviations in its posture. A guiding wheel is employed to rectify its roll angle and ensure the precise spatial positioning of defects both inside and outside the pipeline, as detected by the radar antenna. By analyzing its deflection factors and correction trajectories, the intelligent correction control of the pipeline ground-penetrating radar falls into the realm of nonlinear multi-constraint optimization. Consequently, a time-series-based correction angle prediction algorithm is proposed. The application of the long short-term memory (LSTM) deep learning model facilitates the prediction of correction angles and torque for the guiding wheel. This study compares the performance of LSTM with an autoregressive integrated moving average model under identical dataset conditions. The subsequent findings reveal a reduction of 4.11° and 8.25 N·m in mean absolute error, and a decrease of 10.66% and 7.27% in mean squared error for the predicted correction angles and torques, respectively. These outcomes are achieved utilizing the three-channel drainage pipeline ground-penetrating radar device with top antenna operating at 1.2 GHz and left/right antennas at 750 MHz. The LSTM prediction model intelligently corrects its posture. Experimental results demonstrate an average correction speed of 5 s and an average angular error of ±1°. It is verified that the model can correct its attitude in real-time with small errors, thereby enhancing the accuracy of ground-penetrating radar antennas in locating pipeline defects.

Aeroengine Working Condition Recognition Based on MsCNN-BiLSTM.

Aeroengine working condition recognition is a pivotal step in engine fault diagnosis. Currently, most research on aeroengine condition recognition focuses on the stable condition. To identify the aeroengine working conditions including transition conditions and better achieve the fault diagnosis of engines, a recognition method based on the combination of multi-scale convolutional neural networks (MsCNNs) and bidirectional long short-term memory neural networks (BiLSTM) is proposed. Firstly, the MsCNN is used to extract the multi-scale features from the flight data. Subsequently, the spatial and channel weights are corrected using the weight adaptive correction module. Then, the BiLSTM is used to extract the temporal dependencies in the data. The Focal Loss is used as the loss function to improve the recognition ability of the model for confusable samples. L2 regularization and DropOut strategies are employed to prevent overfitting. Finally, the established model is used to identify the working conditions of an engine sortie, and the recognition results of different models are compared. The overall recognition accuracy of the proposed model reaches over 97%, and the recognition accuracy of transition conditions reaches 94%. The results show that the method based on MsCNN-BiLSTM can effectively identify the aeroengine working conditions including transition conditions accurately.

An Enhanced Full-Form Model-Free Adaptive Controller for SISO Discrete-Time Nonlinear Systems.

This study focuses on the full-form model-free adaptive controller (FFMFAC) for SISO discrete-time nonlinear systems, and proposes enhanced FFMFAC. The proposed technique design incorporates long short-term memory neural networks (LSTMs) and fuzzy neural networks (FNNs). To be more precise, LSTMs are utilized to adjust vital parameters of the FFMFAC online. Additionally, due to the high nonlinear approximation capabilities of FNNs, pseudo gradient (PG) values of the controller are estimated online. EFFMFAC is characterized by utilizing the measured I/O data for the online training of all introduced neural networks and does not involve offline training and specific models of the controlled system. Finally, the rationality and superiority are verified by two simulations and a supporting ablation analysis. Five individual performance indices are given, and the experimental findings show that EFFMFAC outperforms all other methods. Especially compared with the FFMFAC, EFFMFAC reduces the RMSE by 21.69% and 11.21%, respectively, proving it to be applicable for SISO discrete-time nonlinear systems.

Monitoring Mixing Processes Using Ultrasonic Sensors and Machine Learning.

Mixing is one of the most common processes across food, chemical, and pharmaceutical manufacturing. Real-time, in-line sensors are required for monitoring, and subsequently optimising, essential processes such as mixing. Ultrasonic sensors are low-cost, real-time, in-line, and applicable to characterise opaque systems. In this study, a non-invasive, reflection-mode ultrasonic measurement technique was used to monitor two model mixing systems. The two systems studied were honey-water blending and flour-water batter mixing. Classification machine learning models were developed to predict if materials were mixed or not mixed. Regression machine learning models were developed to predict the time remaining until mixing completion. Artificial neural networks, support vector machines, long short-term memory neural networks, and convolutional neural networks were tested, along with different methods for engineering features from ultrasonic waveforms in both the time and frequency domain. Comparisons between using a single sensor and performing multisensor data fusion between two sensors were made. Classification accuracies of up to 96.3% for honey-water blending and 92.5% for flour-water batter mixing were achieved, along with R2 values for the regression models of up to 0.977 for honey-water blending and 0.968 for flour-water batter mixing. Each prediction task produced optimal performance with different algorithms and feature engineering methods, vindicating the extensive comparison between different machine learning approaches.

Deep Neural Network Based Predictions of Protein Interactions Using Primary Sequences.

Machine learning based predictions of protein⁻protein interactions (PPIs) could provide valuable insights into protein functions, disease occurrence, and therapy design on a large scale. The intensive feature engineering in most of these methods makes the prediction task more tedious and trivial. The emerging deep learning technology enabling automatic feature engineering is gaining great success in various fields. However, the over-fitting and generalization of its models are not yet well investigated in most scenarios. Here, we present a deep neural network framework (DNN-PPI) for predicting PPIs using features learned automatically only from protein primary sequences. Within the framework, the sequences of two interacting proteins are sequentially fed into the encoding, embedding, convolution neural network (CNN), and long short-term memory (LSTM) neural network layers. Then, a concatenated vector of the two outputs from the previous layer is wired as the input of the fully connected neural network. Finally, the Adam optimizer is applied to learn the network weights in a back-propagation fashion. The different types of features, including semantic associations between amino acids, position-related sequence segments (motif), and their long- and short-term dependencies, are captured in the embedding, CNN and LSTM layers, respectively. When the model was trained on Pan's human PPI dataset, it achieved a prediction accuracy of 98.78% at the Matthew's correlation coefficient (MCC) of 97.57%. The prediction accuracies for six external datasets ranged from 92.80% to 97.89%, making them superior to those achieved with previous methods. When performed on Escherichia coli, Drosophila, and Caenorhabditis elegans datasets, DNN-PPI obtained prediction accuracies of 95.949%, 98.389%, and 98.669%, respectively. The performances in cross-species testing among the four species above coincided in their evolutionary distances. However, when testing Mus Musculus using the models from those species, they all obtained prediction accuracies of over 92.43%, which is difficult to achieve and worthy of note for further study. These results suggest that DNN-PPI has remarkable generalization and is a promising tool for identifying protein interactions.

Tuning attention based long-short term memory neural networks for Parkinson's disease detection using modified metaheuristics.

Parkinson's disease (PD) is a progressively debilitating neurodegenerative disorder that primarily affects the dopaminergic system in the basal ganglia, impacting millions of individuals globally. The clinical manifestations of the disease include resting tremors, muscle rigidity, bradykinesia, and postural instability. Diagnosis relies mainly on clinical evaluation, lacking reliable diagnostic tests and being inherently imprecise and subjective. Early detection of PD is crucial for initiating treatments that, while unable to cure the chronic condition, can enhance the life quality of patients and alleviate symptoms. This study explores the potential of utilizing long-short term memory neural networks (LSTM) with attention mechanisms to detect Parkinson's disease based on dual-task walking test data. Given that the performance of networks is significantly inductance by architecture and training parameter choices, a modified version of the recently introduced crayfish optimization algorithm (COA) is proposed, specifically tailored to the requirements of this investigation. The proposed optimizer is assessed on a publicly accessible real-world clinical gait in Parkinson's disease dataset, and the results demonstrate its promise, achieving an accuracy of 87.4187 % for the best-constructed models.

A lightweight intrusion detection method for IoT based on deep learning and dynamic quantization.

Intrusion detection ensures that IoT can protect itself against malicious intrusions in extensive and intricate network traffic data. In recent years, deep learning has been extensively and effectively employed in IoT intrusion detection. However, the limited computing power and storage space of IoT devices restrict the feasibility of deploying resource-intensive intrusion detection systems on them. This article introduces the DL-BiLSTM lightweight IoT intrusion detection model. By combining deep neural networks (DNNs) and bidirectional long short-term memory networks (BiLSTMs), the model enables nonlinear and bidirectional long-distance feature extraction of complex network information. This capability allows the system to capture complex patterns and behaviors related to cyber-attacks, thus enhancing detection performance. To address the resource constraints of IoT devices, the model utilizes the incremental principal component analysis (IPCA) algorithm for feature dimensionality reduction. Additionally, dynamic quantization is employed to trim the specified cell structure of the model, thereby reducing the computational burden on IoT devices while preserving accurate detection capability. The experimental results on the benchmark datasets CIC IDS2017, N-BaIoT, and CICIoT2023 demonstrate that DL-BiLSTM surpasses traditional deep learning models and cutting-edge detection techniques in terms of detection performance, while maintaining a lower model complexity.

A two-dimensional hydrodynamics prediction framework for mantle-undulated propulsion robot using multiple proper orthogonal decomposition and long short term memory neural network.

In this paper, a deep learning based framework has been developed to predict hydrodynamic forces on a mantle-undulated propulsion robot (MUPRo). A multiple proper orthogonal decomposition (MPOD) algorithm has been proposed to efficiently identify fluid features near the undulating mantle of the MUPRo globally and locally. The results indicate that theL2error of the solution states near the undulating boundary of the proposed MPOD algorithm converges almost linearly to 0.2%. Furthermore, a hydrodynamics prediction framework has been developed based on the proposed MPOD algorithm, where a long short-term memory neural network predicts the temporal coefficients of the MPOD spatial modes. The developed framework achieves economical and reliable predictions of hydrodynamic forces acting on the undulating boundary compared to simulations and experiments. Moreover, theL2error of the developed framework is one to two orders of magnitude lower than that of the frameworks based on the classical POD algorithm when the degrees of freedom are consistent. Finally, the reliability of the proposed MPOD-NIROM is discussed through an offline parameter planning case of an aquatic-inspired robot. The model presented in this paper can provide support for the offline parameter planning of aquatic-inspired robots.

Predicting Analyte Concentrations from Electrochemical Aptasensor Signals Using LSTM Recurrent Networks.

Nanomaterial-based aptasensors are useful devices capable of detecting small biological species. Determining suitable signal processing methods can improve the identification and quantification of target analytes detected by the biosensor and consequently improve the biosensor's performance. In this work, we propose a data augmentation method to overcome the insufficient amount of available original data and long short-term memory (LSTM) to automatically predict the analyte concentration from part of a signal registered by three electrochemical aptasensors, with differences in bioreceptors, analytes, and the signals' lengths for specific concentrations. To find the optimal network, we altered the following variables: the LSTM layer structure (unidirectional LSTM (LSTM) and bidirectional LSTM (BLSTM)), optimizers (Adam, RMSPROP, SGDM), number of hidden units, and amount of augmented data. Then, the evaluation of the networks revealed that the highest original data accuracy increased from 50% to 92% by exploiting the data augmentation method. In addition, the SGDM optimizer showed a lower performance prediction than that of the ADAM and RMSPROP algorithms, and the number of hidden units was ineffective in improving the networks' performances. Moreover, the BLSTM nets showed more accurate predictions than those of the ULSTM nets on lengthier signals. These results demonstrate that this method can automatically detect the analyte concentration from the sensor signals.

Training LSTM-neural networks on early warning signals of declining cooperation in simulated repeated public good games.

We present results of attempts to expand and enhance the predictive power of Early Warning Signals (EWS) for Critical Transitions (Scheffer et al. 2009) through the deployment of a Long-Short-Term-Memory (LSTM) Neural Network on agent-based simulations of a Repeated Public Good Game, which due to positive feedbacks on experience and social entrainment transits abruptly from majority cooperation to majority defection and back. Our method extension is inspired by several known deficiencies of EWS and by lacking possibilities to consider micro-level interaction in the so far primarily used simulation methods. We find that•The method is applicable to agent-based simulations (as an extension of equation-based methods).•The LSTM yields signals of imminent transitions that can complement statistical indicators of EWS.•The less tensely connected part of an agent population could take a larger role in causing a tipping than the well-connected part.