Delay Neural Network Research Articles

Leveraging Singular Spectrum Analysis and Time Delay Neural Network for Improved Potato Price Forecasting

Leveraging Singular Spectrum Analysis and Time Delay Neural Network for Improved Potato Price Forecasting

A matrix-separation-based integral inequality for aperiodic sampled-data synchronization of delayed neural networks considering communication delay

This paper achieves the synchronization of delayed neural networks (DNNs) considering aperiodic sampled-data control and communication delay. First of all, based on the master-slave DNNs with aperiodic sampling synchronization controller, a synchronization error system is constructed. Then, an augmented functional containing both the error state and its derivative is constructed. Compared with the existing researches, the augmented functional introduces more cross information of error states to the criterion. Next, an integral inequality based on the separation of internal integral variable and matrix is developed. Compared to the inequalities that treat the internal variable and the matrix as unified ones, the developed inequality provides a tighter estimate of the derivative of the augmented functional. On this basis, a criterion with less conservative is developed for the aperiodic sampled-data synchronization of DNNs considering communication delay. Finally, to indicate the superiority of the developed method on improving the acceptable sampling upper bound of synchronization, three numerical examples are provided.

Adaptive QSMO-Based Sensorless Drive for IPM Motor with NN-Based Transient Position Error Compensation

In commercial electrical equipment, the popular sensorless drive scheme for the interior permanent magnet synchronous motor, based on the quasi-sliding mode observer (QSMO) and phase-locked loop (PLL), still faces challenges such as position errors and limited applicability across a wide speed range. To address these problems, this paper analyzes the frequency domain model of the QSMO. A QSMO-based parameter adaptation method is proposed to adjust the boundary layer and widen the speed operating range, considering the QSMO bandwidth. A QSMO-based phase lag compensation method is proposed to mitigate steady-state position errors, considering the QSMO phase lag. Then, the PLL model is analyzed to select the estimated speed difference for transient position error compensation. Specifically, a transient position error compensator based on a feedback time delay neural network (FB-TDNN) is proposed. Based on the back propagation learning algorithm, the specific structure and optimal parameters of the FB-TDNN are determined during the offline training process. The proposed parameter adaptation method and two position error compensation methods were validated through simulations in simulated wide-speed operation scenarios, including sudden speed changes. Overall, the proposed scheme fully mitigates steady-state position errors, substantially mitigates transient position errors, and exhibits good stability across a wide speed range.

Data-driven neural network model for maximum speed profile based on driver risk

In the field of vehicle control systems, the primary objective is to ensure the safety of road users. The intricate dynamics of road vehicles necessitate a high level of precision. Vehicle safety encompasses a multitude of considerations, including vehicle trajectory, prevailing traffic conditions, road structure, and meteorological factors. This study employs an Artificial Neural Network (ANN) trained with human driver data using SCANeR Studio software to evaluate the risk for the driver. The risk has been defined as a five-level parameter, which depends on the potential danger of a situation, where speed and direction play a crucial role. The system incorporates a simulator, an ANN, and a display interface to present the surroundings and communicate important information to the driver. This research employs a simulated driving scenario comprising a multi-lane roundabout with vehicles travelling in different directions to simulate real-world challenges. Risk estimation is achieved through a Time Delay Neural Network (TDNN) trained with various information about the environment in relation to the driven vehicle. The research employs a Jackknife technique for overall evaluation and introduces an adaptive algorithm for speed limit setting. The findings demonstrate the stability, generality, and practical applicability of the ANN in enhancing road safety.

A Degree-Dependent Polynomial-Based Reciprocally Convex Matrix Inequality and Its Application to Stability Analysis of Delayed Neural Networks.

In this article, several improved stability criteria for time-varying delayed neural networks (DNNs) are proposed. A degree-dependent polynomial-based reciprocally convex matrix inequality (RCMI) is proposed for obtaining less conservative stability criteria. Unlike previous RCMIs, the matrix inequality in this article produces a polynomial of any degree in the time-varying delay, which helps to reduce conservatism. In addition, to reduce the computational complexity caused by dealing with the negative definite of the high-degree terms, an improved lemma is presented. Applying the above matrix inequalities and improved negative definiteness condition helps to generate a more relaxed stability criterion for analyzing time-varying DNNs. Two examples are provided to illustrate this statement.

Meander Structure Analysis Techniques Using Artificial Neural Networks

Typically, analyses of meander structures (MSs) for transfer characteristics are conducted using specialized commercial software based on numerical methods. However, these methods can be time-consuming, particularly when a researcher is seeking to perform a preliminary study of the designed structures. This study aims to explore the application of neural networks in the design and analysis of meander structures. Three different feedforward neural network (FFNN), time delay neural network (TDNN), and convolutional neural network (CNN) techniques were investigated for the analysis and design of the meander structures in this article. The geometric dimensions or top-view images of 369 different meander structures were used for training an FFNN, TDNN, and CNN. The investigated networks were designed to predict such electrodynamic parameters as the delay time (td), reflection coefficient (S11), and transmission coefficient (S21) in the 0–10 GHz frequency band. A sufficiently low mean absolute error (MAE) was achieved with all three methods for the analysis of MSs. Using an FFNN, the characteristic td was predicted with a 3.3 ps average MAE. The characteristic S21 was predicted with a 0.64 dB average MAE, and S11 was predicted with a 2.47 dB average MAE. The TDNN allowed the average MAEs to be reduced to 0.9 ps, 0.11 dB, and 1.63 dB, respectively. Using a CNN, the average MAEs were 27.5 ps, 0.44 dB, and 1.36 dB, respectively. The use of neural networks has allowed accelerating the analysis procedure from approximately 120 min on average to less than 5 min.

Novel Caputo fractional differential approach and hybrid control scheme for finite‐time synchronization of nonidentical delayed reaction–diffusion neural networks

AbstractThis work focuses on the finite‐time synchronization (F‐TS) for nonidentical delayed neural networks (DNNs) including Caputo fractional partial differential operator and reaction–diffusion terms. A novel F‐TS lemma is proposed by constructing a new Caputo differential inequality. Applying the new lemma and designing two hybrid controllers with time delay and the sign function, the F‐TS criteria on the introduced Caputo reaction–diffusion nonidentical DNNs under the Neumann boundary condition are derived by making use of the Filippov differential inclusion, Green's theorem, and fractional Razumikhin theorem. The conditions are expressed through the algebraic inequality which can greatly decrease the computation in checking the F‐TS performance. Moreover, the validity and correctness of the F‐TS results are illustrated by selecting various orders, spatial positions, and diffusion parameters.

Intermittent event-triggered control for exponential synchronization of delayed neural networks on time Scales

This paper studies the exponential synchronization of delayed neural networks (DNNs) on time scales using the intermittent event-triggered control (IETC) method. Initially, considering the time scale situation, an IETC that merges intermittent control and event-triggered control is introduced, and a new differential inequality is developed. Subsequently, an exponential synchronization criterion is established based on the Lyapunov function, the proposed differential inequality and the time scale theory, applicable to continuous, discrete and hybrid time domains. Furthermore, under the event-triggered condition, it is demonstrated that the lower bound of each event-triggered interval exceeds a positive constant, thereby preventing the occurrence of Zeno’s behavior. Finally, the approach efficacy is validated through numerical examples.

A wave delay neural network for solving label-constrained shortest route query on time-varying communication networks.

The focus of the research is on the label-constrained time-varying shortest route query problem on time-varying communication networks. To the best of our knowledge, research on this issue is still relatively limited, and similar studies have the drawbacks of low solution accuracy and slow computational speed. In this study, a wave delay neural network (WDNN) framework and corresponding algorithms is proposed to effectively solve the label-constrained time-varying shortest routing query problem. This framework accurately simulates the time-varying characteristics of the network without any training requirements. WDNN adopts a new type of wave neuron, which is independently designed and all neurons are parallelly computed on WDNN. This algorithm determines the shortest route based on the waves received by the destination neuron (node). Furthermore, the time complexity and correctness of the proposed algorithm were analyzed in detail in this study, and the performance of the algorithm was analyzed in depth by comparing it with existing algorithms on randomly generated and real networks. The research results indicate that the proposed algorithm outperforms current existing algorithms in terms of response speed and computational accuracy.

Machine Learning and Hybrid Model Building Mechanism for Forecasting Agricultural Market Behaviour: A Case Study of Vegetable Crops

A strategic market intelligence mechanism is essential for stakeholders to make well-informed decisions on production, pricing, and marketing strategies; however, the complex behaviour of agricultural markets, coupled with the presence of non-stationarity, non-normality, and non-linearity in the arrivals and prices of perishable agricultural commodities pose challenges for traditional statistical models in modelling and forecasting. This study sought to overcome these challenges by employing machine learning and hybrid models for vegetable crops. Machine learning techniques, known for their flexibility with non-stationary, non-normal, and non-linear data, were showcased in the empirical illustration, highlighting their effectiveness. The trends of arrivals and prices were found to be significantly positive (p<0.01) for tomato and capsicum. Time delay neural network (TDNN) and hybrid models outperformed the seasonal ARIMA models based on lower RMSE and MAPE for the modelling and forecasting. Utilizing the best-fit models, the prices of tomato and capsicum were forecasted to be highest during June 2024 with ?2267 and 2623 per q. This study contributed to the growing body of applications of machine learning in agricultural systems. It will help develop a robust market intelligence mechanism to address uncertainties in agricultural markets.

Stability analysis of switched stochastic time‐varying delay neural networks with asynchronous switching aperiodic intermittent control

AbstractThis paper is dedicated to studying the mean square exponential stability of switched stochastic time‐varying delay neural networks with asynchronous switching aperiodic intermittent control. By utilizing the Halanay‐type inequality and designing appropriate asynchronous switching aperiodic intermittent control, several novel stability criteria of switched stochastic delay neural networks with asynchronous aperiodic intermittent control are presented under the Lyapunov function method and Lyapunov functional method. Among them, asynchronous switching aperiodic intermittent controllers are more practical and flexible compared to state feedback controllers. Finally, we provide two examples to verify the effectiveness of the theoretical results.

Comparative Study of EMD based Modelling Techniques for Improved Agricultural Price Forecasting

Forecasting agricultural commodity prices is regarded as a challenging task due to its non-linear and non-stationary nature. As agriculture productionis highly reliant on various biological and agro-meteorological factors, traditional smoothing techniques as well as statistical models often fail tomodel such series satisfactorily. To capture such complex patterns effectively, different data-driven and self-adaptive techniques have been developedtime-to-time. Against this backdrop, in this paper, we have assessed the suitability of empirical mode decomposition (EMD)-based neural network andsupport vector regression (SVR) approaches for forecasting wholesale prices of three major potato markets namely, Agra, Bangalore, and Mumbai.As the benchmark models, autoregressive integrated moving average (ARIMA), time delay neural network (TDNN) and SVR models have beenemployed for the comparative evaluation. The experimental results clearly reveal the comparative superiority of the EMD-SVR model for the Agraand Bangalore markets and the EMD-TDNN model for the Mumbai market in terms of root mean squared error values and turning point predictions.Moreover, all the EMD-based models have performed better than the other competing models.

Constructing selection hyper-heuristics for open vehicle routing with time delay neural networks using multiple experts

Hyper-heuristics are general purpose search methods for solving computationally difficult problems. A selection hyper-heuristic is composed of two key components: a heuristic selection method and move acceptance criterion. Under an iterative single-point search framework, a solution is modified by selecting and applying a predefined low-level heuristic, with a decision then taken to accept or reject the resulting solution. Designing a selection hyper-heuristic is an extremely challenging task. In this study, we investigate computer-aided design of a selection hyper-heuristic for the open vehicle routing problem. A time delay neural network is used as an offline apprenticeship learning method. Our approach first observes the search behaviour of multiple expert human-designed selection hyper-heuristics on a selected sample of training instances, before automatically generating a selection hyper-heuristic capable of solving unseen instances effectively. The proposed approach is tested on open vehicle routing problem instances of different sizes to examine the performance and generality of the selection hyper-heuristics generated. Improved performance is demonstrated over a set of well-known benchmarks from the literature when compared to using the existing expert systems directly.

Modeling nonlinear dynamic objects using pre-trained time delay neural networks

The work is devoted to resolving the contradiction between the accuracy of modelling nonlinear dynamics and the speed of model construction under conditions of limited computing resources. The purposeof the work is to reduce the time for building time delay neural networks while ensuring a given accuracy in the tasks of identifying nonlinear dynamic objects with continuous characteristics. This goal is achieved by developing a method for pre-training neural networks that reflect the basic characteristics of the subject area. The scientific novelty of the work is the development of a method for identifying nonlinear dynamic objectsin the form of time delay neural networks based on the use of a set of basic pre-trained neural networks that reflect the typical properties of the subject area. In contrast to the traditional approach to pre-training, the developed method allows building models of lower complexity.A formal criterion is proposed for determining the moment of termination of the neural network pre-training, the use of which allows avoiding retraining of the base model and ensuring a significant reduction in the model training time on the target data set. The practical utility of the work lies in the development of an algorithm for the method of pre-training time delay neural networks in the tasks of identifying nonlinear dynamic objects with continuous characteristics, which allows to significantlyreduce the training time of neural networks without losing model accuracy. The value of this study is todetermine the area of effective use of the proposed method, namely, when the general and target datasets do not have significant differences and the target dataset is of sufficient size to reflect the properties of the research object.

Asymptotic stability of nonlinear fractional delay differential equations with α ∈ (1, 2): An application to fractional delay neural networks.

We introduce a theorem on linearized asymptotic stability for nonlinear fractional delay differential equations (FDDEs) with a Caputo order α∈(1,2), which can be directly used for fractional delay neural networks. It relies on three technical tools: a detailed root analysis for the characteristic equation, estimation for the generalized Mittag-Leffler function, and Lyapunov's first method. We propose coefficient-type criteria to ensure the stability of linear FDDEs through a detailed root analysis for the characteristic equation obtained by the Laplace transform. Further, under the criteria, we provide a wise expression of the generalized Mittag-Leffler functions and prove their polynomial long-time decay rates. Utilizing the well-established Lyapunov's first method, we establish that an equilibrium of a nonlinear Caputo FDDE attains asymptotically stability if its linearization system around the equilibrium solution is asymptotically stable. Finally, as a by-product of our results, we explicitly describe the asymptotic properties of fractional delay neural networks. To illustrate the effectiveness of our theoretical results, numerical simulations are also presented.

Whisper40: A Multi-Person Chinese Whisper Speaker Recognition Dataset Containing Same-Text Neutral Speech

Whisper speaker recognition (WSR) has received extensive attention from researchers in recent years, and it plays an important role in medical, judicial, and other fields. Among them, the establishment of a whisper dataset is very important for the study of WSR. However, the existing whisper dataset suffers from the problems of a small number of speakers, short speech duration, and lack of neutral speech with the same-text as the whispered speech in the same dataset. To address this issue, we present Whisper40, a multi-person Chinese WSR dataset containing same-text neutral speech spanning around 655.90 min sourced from volunteers. In addition, we use the current state-of-the-art speaker recognition model to build a WSR baseline system and combine the idea of transfer learning for pre-training the speaker recognition model using neutral speech datasets and transfer the empirical knowledge of specific network layers to the WSR system. The Whisper40 and CHAINs datasets are then used to fine-tune the model with transferred specific layers. The experimental results show that the Whisper40 dataset is practical, and the time delay neural network (TDNN) model performs well in both the same/cross-scene experiments. The equal error rate (EER) of Chinese WSR after transfer learning is reduced by 27.62% in comparison.

MECOS: A bilingual Manipuri–English spontaneous code-switching speech corpus for automatic speech recognition

In this study, we introduce a new code-switched speech database with 57h of Manipuri–English annotated spontaneous speech. Manipuri is an official language of India and is primarily spoken in the north–eastern Indian state of Manipur. Most Manipur native speakers today are bilingual and frequently use code switching in everyday discussions. By carefully assessing the amount of code-switched speech in each video, recordings from YouTube are gathered. 21,339 utterances and 291,731 instances of code switching are present in the database. Given the code-switching nature of the data, a proper annotation procedure is used, and the data are manually annotated using the Meitei Mayek unicode font and the roman alphabets for Manipuri and English, respectively. The transcription includes the information of the speakers, non-speech information, and the corresponding annotation. The aim of this research is to construct an automatic speech recognition (ASR) system as well as offer a thorough analysis and details of the speech corpus. We believe that our research is the first to use an ASR system for Manipuri–English code-switched speech. To evaluate the performance, ASR systems based on hybrid deep neural network and hidden Markov model (DNN–HMM), time delay neural network (TDNN), hybrid time delay neural network and long short-term memory (TDNN–LSTM) and three end-to-end (E2E) models i.e. hybrid connectionist temporal classification and attention model (CTC-Attention), Conformer, wav2vec XLSR are developed for Manipuri–English language. In comparison to other models, pure TDNN produces outcomes that are clearly superior.

Mixed-Delay-Based Augmented Functional for Sampled-Data Synchronization of Delayed Neural Networks With Communication Delay.

The synchronization control for delayed neural networks (DNNs) via a sampled-data controller considering communication delay is studied by input delay approach. Although few scholars have put forward the coexistence of transmission delay and communication delay in this problem, no report has clarified the interaction between transmission delay and communication delay. Also, the time-squared terms are underutilized. Thus, a novel augmented Lyapunov functional, which consists of a mixed-delay-based augmented part and a time-squared two-sided looped part, is proposed to fill this gap. In the mixed-delay-based augmented part, not only the information of transmission delay and communication delay themselves, but also the interaction between those two delays is considered. Time-dependent quadratic terms as well as the sampling integral states are introduced in the two-sided looped part, so that more characteristic information of the sampling pattern is encompassed and the relationship of the states at the sampling instant is enhanced. Then, this novel augmented functional is applied to the synchronization control of DNNs. A less conservative synchronization criterion is obtained in the form of linear matrix inequalities. A numerical example illustrates the validity and superiority of the presented synchronization criterion.

Joint speaker encoder and neural back-end model for fully end-to-end automatic speaker verification with multiple enrollment utterances

Conventional automatic speaker verification systems can usually be decomposed into a front-end model such as time delay neural network (TDNN) for extracting speaker embeddings and a back-end model such as statistics-based probabilistic linear discriminant analysis (PLDA) or neural network-based neural PLDA (NPLDA) for similarity scoring. However, the sequential optimization of the front-end and back-end models may lead to a local minimum, which theoretically prevents the whole system from achieving the best optimization. Although some methods have been proposed for jointly optimizing the two models, such as the generalized end-to-end (GE2E) model and NPLDA E2E model, most of these methods have not fully investigated how to model the intra-relationship between multiple enrollment utterances. In this paper, we propose a new E2E joint method for speaker verification especially designed for the practical scenario of multiple enrollment utterances. To leverage the intra-relationship among multiple enrollment utterances, our model comes equipped with frame-level and utterance-level attention mechanisms. Additionally, focal loss is utilized to balance the importance of positive and negative samples within a mini-batch and focus on the difficult samples during the training process. We also utilize several data augmentation techniques, including conventional noise augmentation using MUSAN and RIRs datasets and a unique speaker embedding-level mixup strategy for better optimization.

μ-stability and instability of multiple equilibrium points in delayed neural networks with general discontinuous activation functions

This paper proposes a general type of discontinuous activation functions (AFs) and studies the μ-stability of multiple equilibrium points (EPs) in delayed neural networks (DNNs). By judging 4n algebraic inequalities and the nonsingularity of an M-matrix, DNNs with the general discontinuous AFs can be shown to have 5n EPs, therein 3n EPs are locally μ-stable. Moreover, these 3n EPs are located at the points of continuity of the AF. Compared with the existing general continuous AF, DNNs with the general discontinuous AF can have more total/stable EPs. Hence, DNNs with the general discontinuous AF could store a much larger number of memory patterns if they are applied to associative memory. In addition, the attraction basin (AB) of each stable EP in DNNs is estimated. Two numerical examples are shown to testify the validity of the obtained results.