Narrow Results

Electric power resources are essential for the efficient and orderly development of society. Accurate power load forecasting is a key driver for the low-carbon upgrade of power systems. Traditional forecasting methods often struggle to capture long-term dependencies. Additionally, extracting complex nonlinear features from data remains a significant challenge, making it challenging to meet the accuracy demands of modern power systems. Besides, current deep learning-based forecasting methods cannot simulate multi-granularity power load data. To address these challenges, this paper presents a Generative Pre-trained Transformer model, GPT4PLTS, designed for power data simulation and fine-grained power load forecasting. The model leverages the Transformer architecture, incorporating the first six layers of the GPT decoder structure. It utilizes a multi-head attention mechanism to extract temporal features and includes a time alignment layer to maintain the sequence of time-series data, addressing both short-term and long-term dependencies. Extensive experiments are conducted on load observations from 2000 enterprises. The results demonstrate that GPT4PLTS achieves high accuracy in data simulation and forecasts across different time granularities, particularly excelling in short and medium-term predictions. Future research could focus on optimizing the model structure to enhance the model’s generalization ability.

Baidu search engine is the most common one adopted by Chinese Internet users, and Baidu index provides a platform to capture the behaviors of massive users on Baidu, which is one important statistical tool to mine the Internet users’ behaviors and characteristics in China. Here, we utilize the Baidu index data on greenhouse gas from January 1, 2011 to November 29, 2019, to perform the related statistical analyses at first. Then, on the basis of Baidu index time series data, the corresponding network is constructed by use of the visibility graph method. Finally, the topology of the generated network is analyzed from different perspectives. Our results indicate that people’s attention to greenhouse gases obeys the power-law distribution, and we can identify the significant nodes and find some outliers in time series data by use of the topological properties of networks. Taking together, the current model offers a novel means to represent and depict the time series data of Baidu index through the complex network analysis.

Due to non-stationary nature of Indian summer monsoon rainfall (ISMR), analysis of its patterns and behaviors is a very tedious task. Advance prediction and behaviors play a significant role in various domains. Literature review reveals that researchers’ works are limited to design predictive models but not on inherited patterns and behaviors for the ISMR. In this study, a novel method based on the hybridization of two computational techniques, viz., fuzzy and rough sets is proposed for patterns and behaviors. The proposed method initially classifies the information into the four distinct regions, as fuzzy positive region, fuzzy negative region, completely fuzzy region, and gray fuzzy region. Based on four regions, four different patterns of decision rules are explored. Further, a method is discussed to represent such decision rules in terms of graph, which helps to analyze the patterns of ISMR by discovering new knowledge. The proposed method is validated by performing various statistical analyses.

Reservoir computing is a machine learning framework suited for pattern recognition of temporal signals. A reservoir computing system normally consists of a dynamic reservoir and a static readout. Memristive systems and devices with intrinsic nonlinearity are potentially favorable for constructing a reservoir hardware because the role of the reservoir is to nonlinearly map input signals into a higher-dimensional feature space. In this study, we propose a minimal configuration of memristive circuit reservoirs for solving temporal pattern classification problems. We numerically demonstrate how the classification performance of the proposed reservoir computing systems with only two memristive devices depends on the system conditions in waveform and electrocardiogram classification tasks. The signal processing methods used for performance improvement provide useful insights into effective and efficient implementation of memristive reservoir computing hardware.

We propose a hybrid architecture based on a combination of fuzzy systems and polynomial neural networks. The resulting Hybrid Fuzzy Polynomial Neural Networks (HFPNN) dwells on the ideas of fuzzy rule-based computing and polynomial neural networks. The structure of the network comprises of fuzzy polynomial neurons (FPNs) forming the nodes of the first (input) layer of the HFPNN and polynomial neurons (PNs) that are located in the consecutive layers of the network. In the FPN (that forms a fuzzy inference system), the generic rules assume the form "if A then y = P(x)" where A is fuzzy relation in the condition space while P(x) is a polynomial standing in the conclusion part of the rule. The conclusion part of the rules, especially the regression polynomial uses several types of high-order polynomials such as constant, linear, quadratic, and modified quadratic. As the premise part of the rules, both triangular and Gaussian-like membership functions are considered. Each PN of the network realizes a polynomial type of partial description (PD) of the mapping between input and out variables. HFPNN is a flexible neural architecture whose structure is based on the Group Method of Data Handling (GMDH) and developed through learning. In particular, the number of layers of the PNN is not fixed in advance but is generated in a dynamic way. The experimental part of the study involves two representative numerical examples such as chaotic time series and Box-Jenkins gas furnace data.

Globally, people’s health and wealth are affected by the outbreak of the corona virus. It is a virus, which infects from common fever to severe acute respiratory syndrome. It has the potency to transmit from one person to another. It is established that this virus spread is augmenting speedily devoid of any symptoms. Therefore, the prediction of this outbreak situation with mathematical modelling is highly significant along with necessary. To produce informed decisions along with to adopt pertinent control measures, a number of outbreak prediction methodologies for COVID-19 are being utilized by officials worldwide. An effectual COVID-19 outbreaks’ prediction by employing Squirrel Search Optimization Algorithm centric Tanh Multi-Layer Perceptron Neural Network (MLPNN) (SSOA-TMLPNN) along with Auto-Regressive Integrated Moving Average (ARIMA) methodologies is proposed here. Initially, from the openly accessible sources, the input time series COVID-19 data are amassed. Then, pre-processing is performed for better classification outcomes after collecting the data. Next, by utilizing Sine-centered Empirical Mode Decomposition (S-EMD) methodology, the data decomposition is executed. Subsequently, the data are input to the Brownian motion Intense (BI) - SSOA-TMLPNN classifier. In this, the diseased, recovered, and death cases in the country are classified. After that, regarding the time-series data, the corona-virus’s future outbreak is predicted by employing ARIMA. Afterwards, data visualization is conducted. Lastly, to evaluate the proposed model’s efficacy, its outcomes are analogized with certain prevailing methodologies. The obtained outcomes revealed that the proposed methodology surpassed the other existing methodologies.

In light of the existing challenges in capturing short-term fluctuations and achieving accurate predictions in stock market time series data, this research presents the “Dual-Attention MDWT-CVT-LSTM,” a revolutionary financial time series forecasting model. This model makes use of a dual-attention mechanism in conjunction with a Variant Transformation-Gated Long Short-Term Memory (LSTM) network. The approach relies on the Multilevel Discrete Wavelet Transform (MDWT) to separate stock market index sequences into high-frequency and low-frequency components. Furthermore, a transformative gating mechanism is incorporated into the LSTM, featuring fusion gated units to create the VT-LSTM, effectively addressing abrupt short-term information changes. Within the dual-attention network, VT-LSTM is combined with one-dimensional temporal convolution (Conv1D) to extract spatial and temporal features of data at different frequencies. This facilitates the prediction of various sub-sequences, enabling multilevel and multi-path forecasting research. The experimental evaluation uses diverse models on stock market index datasets and individual stock datasets. The results show the superior prediction accuracy of the proposed model compared to alternative methods, underscoring its practical viability in stock market forecasting applications.

Food price forecasts in the agricultural sector have always been a vital matter to a wide variety of market participants. In this work, we approach this forecast problem for the weekly wholesale food price index in the Chinese market during a 10-year period of January 1, 2010–January 3, 2020. To facilitate the analysis, we propose the use of the nonlinear auto-regressive neural network. Technically, we investigate forecast performance, based upon the relative root mean square error (RRMSE) as the evaluation metrics, corresponding to one hundred and twenty settings that cover different algorithms for model estimations, numbers of hidden neurons and delays, and ratios for splitting the data. Our experimental result suggests the construction of the neural network with three delays and 10 hidden neurons, which is trained through the Levenberg–Marquardt algorithm, as the forecast model. It leads to high accuracy and stabilities with the RRMSEs of 1.93% for the training phase, 2.16% for the validation phase, and 1.95% for the testing phase. Comparisons of forecast accuracy between the proposed model and some other machine learning models, as well as traditional time-series econometric models, suggest that our proposed model leads to statistically significant better performance. Our results could benefit different forecast users, such as policymakers and various market participants, in policy analysis and market assessments.

With a tremendous increase in mobile and wearable devices, the study of sensor-based activity recognition has drawn a lot of attention in the past years. In recent years, the applications of Human Activity Recognition are getting more and more attention, especially in eldercare and healthcare as an assistive technology when combined with the Internet of Things. In this paper, we propose three deep learning approaches to improve the accuracy of activity detection on the WISDM dataset. Particularly, we apply a convolutional neural network to extract the interesting features, then we use softmax function, support vector machine, and random forest for classification tasks. The results show that the hybrid algorithm, convolutional neural network combined with the support vector machine, outperforms all the previous methods in classifying every activity. In addition, not only the support vector machine but also the random forest shows better accuracy in classification task than the neural network classification and the former approaches do.

It is true that anomaly detection is an important issue that has had a long history in the research community due to its various applications. Literature has recorded various Artificial Intelligence (AI) techniques that have been applied to detect anomalies without having a priori knowledge about them. Anomaly detection approaches for multivariate time series data have still too many unrealistic assumptions to apply to the industry. Our paper, therefore, proposed a new efficiency approach of anomaly detection for multivariate time series data. We specifically developed a new hybrid approach based on LSTM Autoencoder and Isolation Forest (iForest). This approach enables the advantages in extracting good features of the LSTM Autoencoder and the good performance in anomaly detection problems of the iForest. The results show that our approach leads to the improvement of performance significantly in comparison with the One-Class Support Vector Machine (OCSVM) method. Our approach is implemented on simulated data in the fashion industry (FI).

Customer classification is one of the major tasks in customer relationship management. Customers have both static characteristics and dynamic behavioral features. To apply both kinds of data to conduct comprehensive analysis can enhance the reasonability of customer classification. In this paper, customer dynamic data is clustered using a hybrid genetic algorithm and then is combined with customer static data to give reasonable customer segmentation by using neural network technique. A novel classification method which considers both the static and dynamic data of customers is proposed. Applying the proposed method in a bank's datasets can obviously improve the accuracy of customer classification comparing with traditional methods where only static data is used.