Research
Artificial Intelligence - Deep Learning - Machine Learning -- Control Systems - Power Systems - Disease Modelling & Analysis
Farooq, Junaid, Danish Rafiq, Pantelis R. Vlachas, and Mohammad Abid Bazaz. "RefreshNet: Learning Multiscale Dynamics through Hierarchical Refreshing." arXiv preprint arXiv:2401.13282 (2024)
Abstract: orecasting complex system dynamics, particularly for long-term predictions, is persistently hindered by error accumulation and computational burdens. This study presents RefreshNet, a multiscale framework developed to overcome these challenges, delivering an unprecedented balance between computational efficiency and predictive accuracy. RefreshNet incorporates convolutional autoencoders to identify a reduced order latent space capturing essential features of the dynamics, and strategically employs multiple recurrent neural network (RNN) blocks operating at varying temporal resolutions within the latent space, thus allowing the capture of latent dynamics at multiple temporal scales. The unique “refreshing” mechanism in RefreshNet allows coarser blocks to reset inputs of finer blocks, effectively controlling and alleviating error accumulation. This design demonstrates superiority over existing techniques regarding computational efficiency and predictive accuracy, especially in long-term forecasting. The framework is validated using three benchmark applications: the FitzHugh-Nagumo system, the Reaction-Diffusion equation, and Kuramoto-Sivashinsky dynamics. RefreshNet significantly outperforms state-of-the-art methods in long-term forecasting accuracy and speed, marking a significant advancement in modeling complex systems and opening new avenues in understanding and predicting their behavior.
7
Conference Paper
Junaid Farooq & Mohammad Abid Bazaz. (2023). Multiscale Autoencoder-RNN Architecture for Mitigating Error Accumulation in Long-Term Forecasting. 2023 International Conference on Emerging Techniques in Computational Intelligence (ICETCI), Hyderabad, India, 2023, pp. 271-275, doi: 10.1109/ICETCI58599.2023.10331086. - (IEEE Computational Intelligence Society). [Received Best Presentation Award]
https://ieeexplore.ieee.org/document/10331086
Abstract: Recurrent Neural Networks (RNNs) have emerged as a vital tool for forecasting complex systems, ranging from natural language processing to high-dimensional dynamical systems. However, during deployment, accurate long-term forecasting requires iterative use of autoregressive scenarios, where predictions are fed back to the RNN as new inputs, leading to propagation and accumulation of error over time. This is further compounded when the state distributions vary, leading to inaccuracies in loss measurement. While previous studies have proposed solutions for language processing networks with probabilistic predictions, these approaches are inadequate for more complex systems. Therefore, in this study, we propose a novel approach to address this gap using a multiscale RNN architecture that leverages high-scale networks to reinitialize low-scale networks, thereby effectively mitigating iterative propagation of errors. To reduce the dimensionality, we integrate an auto encoder with the RNNs. Our experiments, using one of the most widely used RNNs, Long Short-Term Memory (LSTM) networks, demonstrate that our technique significantly reduces iterative errors and accurately forecasts long-term high-dimensional systems such as the FitzHugh-Nagumo model. This technique has the potential to enhance a wide range of applications, including robotics, finance, and climate modeling, by enabling precise long-term prediction of high-dimensional systems.
6
Article
Junaid Farooq, & Mohammad Abid Bazaz (2023). Hybrid Deep Neural Network for Data-driven Missile Guidance with Maneuvering Target. Defence Science Journal, 73(5), 602-611. https://doi.org/10.14429/dsj.73.18481
Abstract:
Missile guidance, owing to highly complex and non-linear relative movement between the missile and its target, is a challenging problem. This is further aggravated in case of a maneuvering target which changes its own flight path while attempting to escape the incoming missile. In this study, to achieve computationally superior and accurate missile guidance, a deep learning is employed to propose a self-tuning technique for a fractional-order proportional integral derivative (FOPID) controller of a radar-guided missile chasing an intelligently maneuvering target. A multi-layer two-dimensional architecture is proposed for a deep neural network that combines the prediction feature of recurrent neural networks and estimation feature of feed-forward artificial neural networks. The proposed deep learning based missile guidance scheme is non-intrusive, data-based, and model-free wherein the parameters are optimized on-the-run while predicting the target’s maneuvering tactics to correct for processing time and loop delays of the system. Using deep learning for online optimization with minimal computational burden is the core feature of the proposed technique. Dual-core parallel simulations of missile-target dynamics and the control system were performed to demonstrate superiority of the proposed scheme in feasibility, adaptability, and the ability to effectively minimize the miss-distance in comparison with traditional and neural offline-tuned PID and FOPID based techniques. Compared to state-of-the-art offline-tuned neural control, the miss-distance was reduced by 68.42% for randomly maneuvering targets. Furthermore, a minimum miss-distance of 0.97 m was achieved for intelligently maneuvering targets for which the state-of-the-art method failed to hit the target. Overall, the proposed technique offers a novel approach for addressing the challenges of missile guidance in a computationally efficient and effective manner.
Danish Rafiq, Junaid Farooq, & Mohammad Abid Bazaz .(2022) Synergistic use of Intrusive and Non-intrusive Model Order Reduction Techniques for Dynamical Power Grids. International Journal of Electrical Power & Energy Systems, 138, 107908, doi.org/10.1016/j.ijepes.2021.107908
Abstract:
This manuscript combines the recently developed nonlinear moment-matching (NLMM) technique with dynamic mode decomposition (DMD) to obtain a simulation-free reduction framework for power systems. Unlike the conventional model reduction methods for power systems, where the external area is linearized, we consider the nonlinear effective network (EN) and the synchronous motor (SM) power grid models. First, the reduced system is constructed from the solution of the underlying approximate Sylvester equation by exciting the system with user-defined inputs from a representative exogenous system. Then, a non-intrusive reduction is performed using DMD to approximate the nonlinear function via Koopman modes in an equationfree manner. The advantage is that a ‘‘simulation-free’’ nonlinear model order reduction framework is obtained to approximate the response of the large-scale power grid models. Finally, we substantiate our observations using numerical simulations of reduced EN and SM models of the IEEE 118 and IEEE 300 bus systems for realistic fault scenarios. Results show that the overall CPU times of the reduced-order models are lowered to half as compared to the original models while maintaining the fidelity. The results are also compared with POD-DEIM for reference.
4
Article
Manzoor A. Wani, Junaid Farooq, & Danish Mushtaq Wani (2021). Risk assessment of COVID-19 pandemic using deep learning model for J&K in India: a district level analysis. Environmental Science and Pollution Research, 1-11. doi.org/10.1007/s11356-021-17046-9
Junaid Farooq & Mohammad Abid Bazaz. (2021). Deep Learning for Self-Tuning of Control Systems. 2021 International Conference on Emerging Techniques in Computational Intelligence (ICETCI), pp. 6-11, doi.org/10.1109/ICETCI51973.2021.9574048 - (IEEE Computational Intelligence Society). August 25-27, 2021. [Received Best Paper Award]
https://ieeexplore.ieee.org/document/9574048
Abstract: This paper proposes an innovative two-dimensional multi-layered deep neural network (DNN) to achieve adaptive, physics-informed, model-free and data-based control of stochastic, sensitive and highly nonlinear systems. The algorithm design exploits the DNN features of adaptive learning, inference of latent variables and time-series prediction to update the controller parameters on-the-run in real-time while compensating for the loop delays at the same time in addition to the system processing time. The proposed deep learning based self-tuning algorithm (DLSTA) is generic and can be used for online self-tuning of controller parameters in general. For the purpose of this paper, it is applied on the speed control of the brushless DC (BLDC) motor in an electric vehicle (EV) using PID controller on the front-end. The simulation results demonstrate the superiority of the proposed scheme over other conventional tuning methods.
Junaid Farooq & Mohammad Abid Bazaz. (2020). A Deep Learning algorithm for modeling and forecasting of COVID-19 in five worst affected states of India. Alexandria Engineering Journal. doi.org/10.1016/j.aej.2020.09.037
Abstract:
In this paper, deep learning is employed to propose an Artificial Neural Network (ANN) based online incremental learning technique for developing an adaptive and non-intrusive analytical model of Covid-19 pandemic to analyze the temporal dynamics of the disease spread. The model is able to intelligently adapt to new ground realities in real-time eliminating the need to retrain the model from scratch every time a new data set is received from the continuously evolving training data. The model is validated with the historical data and a forecast of the disease spread for 30-days is given in the five worst affected states of India.
Junaid Farooq & Mohammad Abid Bazaz. (2020). A novel adaptive deep learning model of Covid-19 with focus on mortality reduction strategies. Chaos, Solitons & Fractals, 138, 110148. doi.org/10.1016/j.chaos.2020.110148
Abstract:
We employ deep learning to propose an Artificial Neural Network (ANN) based and data stream guided real-time incremental learning algorithm for parameter estimation of a non-intrusive, intelligent, adaptive and online analytical model of Covid-19 disease. Modeling and simulation of such problems pose an additional challenge of continuously evolving training data in which the model parameters change over time depending upon external factors. Our main contribution is that in a scenario of continuously evolving training data, unlike typical deep learning techniques, this non-intrusive algorithm eliminates the need to retrain or rebuild the model from scratch every time a new training data set is received. After validating the model, we use it to study the impact of different strategies for epidemic control. Finally, we propose and simulate a strategy of controlled natural immunization through risk-based population compartmentalization (PC) wherein the population is divided in Low Risk (LR) and High Risk (HR) compartments based on risk factors (like comorbidities and age) and subjected to different disease transmission dynamics by isolating the HR compartment while allowing the LR compartment to develop natural immunity. Upon release from the preventive isolation, the HR compartment finds itself surrounded by enough number of immunized individuals to prevent the spread of infection and thus most of the deaths occurring in this group are avoided.
