Neural Network Optimization Algorithm Research Articles

Augmented Physics-Informed Neural Networks and Seahorse Optimization Algorithm for Music Scene Simulation and Emotion Regulation in Virtual Reality Technology

Virtual Reality (VR) technology has shown promise in simulating music scenes for emotion regulation. The challenge is to enhance the realism and interactivity of VR-based music experiences to create a more immersive and emotionally impactful environment for users. This manuscript proposes an augmented physics-informed neural network and seahorse optimization algorithm for music scene simulation and emotion regulation in virtual reality technology (APINN-SHO-MSS-ERVRT). Initially, the data is collected via head-mounted display (HMD) in real time basis. Afterward, the data is fed to data-adaptive Gaussian average filtering (DAGAF) based pre-processing process. In the pre-processing segment, it improves virtual presence. The pre-processing output is given to augmented physics-informed neural network (APINN)and seahorse optimization algorithm (SHO) for effectively classifying the facial expression for fear and null. Therefore, SHO is proposed to enhance weight parameter of augmented physics-informed neural network classifier, which precisely classifies the facial emotion classifier. The proposed method is implemented in MATLAB and evaluated their performance with existing methods. The performance metrics, like accuracy, precision, sensitivity, specificity, computational time and ROC is analysed to the performance of the proposed method. The proposed APINN-SHO-MSS-ERVRT methods of accuracy are provide 90% higher accuracy for fear and 97% higher accuracy for null emotion is analysed when compared with existing methods GERR-NN-AMSVR, ER-VR-EEG and MTS-VRS-CDI. Proposed APINN-SHO-MSS-ERVRT method attains 0.87%, 0.88% and 0.89%higher ROC analysed to the existing methods respectively.

An Assessment of Geological Hazard Management using Dynamically Stabilized Recurrent Neural Network and Beluga Whale Optimization Algorithm

Mountainous region geological hazards are a leading cause of natural disasters, resulting in significant human and economic losses. Regional topography, landforms, lithology, plant life, geological circumstances, and meteorology all have a significant impact on their creation. Gansu, situated in the interior of Northwestern China, features Lanzhou as its capital and primary urban center, positioned in the southeast of the province. Geological risks, particularly landslides, mudslides, and avalanches, present significant challenges to Gansu Province. Consequently, local authorities are actively devising customized strategies to mitigate these hazards and foster sustainable development. In this research work, an Assessment of Geological Hazard Management using Dynamically Stabilized Recurrent Neural Network and Beluga Whale Optimization Algorithm (AGL-HM-DSRNN-BWOA) is proposed. Initially, the input raster data are gathered from the Normalised Difference Vegetation Index (NDVI) dataset. The input raster data is then pre-processed using Adaptive Actor-Critic Bilateral Filter (A2CBF) to reduce noises and increase the overall quality of the raster data. To classify the geological hazards, the pre-processed raster data are fed into a neural network named DSRNN. The geological hazard is accurately categorized as low risk, medium-low risk, medium-high risk, high risk using proposed DSRNN. In general, DSRNN does not express some adaption of optimization strategies for determining optimal parameters to promise exact classification for managing geological hazard by assessment. Therefore, Beluga Whale Optimization Algorithm (BWOA) is proposed to enhance weight parameter of DSRNN classifier, which precisely assess for managing the geological hazards. The efficiency of the proposed AGL-HM-DSRNN-BWOA approach is evaluated using a number of performance criteria, including accuracy, sensitivity, specificity, ROC, mean square error, root mean square error, mean absolute error. The proposed AGL-HM-DSRNN-BWOA method attains 22.36%, 25.42% and 18.17% higher accuracy, 21.26%, 15.42% and 19.27% higher sensitivity, 28.36%, 25.32% and 28.27% higher F-measure compared with existing methods, such as the Risk assessment and its influencing factors examination of geological hazards in typical mountain environment (RA-IFA-GH-TME), Feasibility study of land cover categorization under normalized difference vegetation index for landslide risk assessment(LCC-NDVI-LRA), and Multiple hazard exposure mapping under machine learning for Salzburg, Austria (MH-EM-SSA-ML) respectively.

Early Warning of Financial Risk Management using AGRNN-COA Approach

Financial risk management was used to protect an entity's financial performance and stability, entails detecting, evaluating, and reducing potential risks, such as credit defaults and market swings. It includes tactics including hedging, diversification, and risk limitation. Financial authorities must immediately responsive for financial risk management system that works with the current economic development. This paper proposes a hybrid approach for early pre-alarm warning of financial risk management system and economic development. The proposed hybrid approach is the combined performance of both the Alternating Graph-Regularized Neural Network (AGRNN) and Crayfish Optimization Algorithm (COA).Commonly it is named as AGRNN-COA technique. The major objective of the proposed approach is to give early pre-alarm warning of financial risk management. AGRNN is designed to give early pre-alarm warning of the financial system of the country. The financial risk management from the AGRNN are optimized by using the COA. By then, the proposed model is implemented in the MATLAB/Simulink working platform and the execution is calculated with the present procedures. The proposed method shows better results in all existing like Financial Risk Management Supervised Algorithm (FRM-SVM), Financial Risk Management Deep Neural Network (FRM-DNN) and Financial Risk Management Heap Based Optimization (FRM-HBO). The accuracy level of proposed FRM-AGRNN-COA approach is 98% that is higher than the other existing methods. The specificity and the F-score of the proposed FRM-AGRNN-COA approach is 99% and 97%. The error rate of the proposed FRM-AGRNN-COA approach is 1.8%, which is very less compared to other existing techniques. From the result, it is conclude that the proposed approach based error is less compared to existing techniques.

Application of Clothing Design Based on Computer Vision Technology Using Hybrid Graph Convolution Neural Network and Lotus Effect Optimization Algorithm

Application of Clothing Design Based on Computer Vision Technology Using Hybrid Graph Convolution Neural Network and Lotus Effect Optimization Algorithm

Enhancing urban blue-green landscape quality assessment through hybrid genetic algorithm-back propagation (GA-BP) neural network approach: a case study in Fucheng, China

This study employs an artificial neural network optimization algorithm, enhanced with a Genetic Algorithm-Back Propagation (GA-BP) network, to assess the service quality of urban water bodies and green spaces, aiming to promote healthy urban environments. From an initial set of 95 variables, 29 key variables were selected, including 17 input variables, such as water and green space area, population size, and urbanization rate, six hidden layer neurons, such as patch number, patch density, and average patch size, and one output variable for the comprehensive value of blue-green landscape quality. The results indicate that the GA-BP network achieves an average relative error of 0.94772%, which is superior to the 1.5988% of the traditional BP network. Moreover, it boasts a prediction accuracy of 90% for the comprehensive value of landscape quality from 2015 to 2022, significantly outperforming the BP network’s approximate 70% accuracy. This method enhances the accuracy of landscape quality assessment but also aids in identifying crucial factors influencing quality. It provides scientific and objective guidance for future urban landscape structure and layout, contributing to high-quality urban development and the creation of exemplary living areas.

Comparative Performance Evaluation of ANN and Hybrid ANN-HBA Models for Temporal and Spatial Modeling of Precipitation in East Azarbaijan Province of Iran

In this study, temporal and spatial modeling of precipitation was performed in East Azarbaijan province of Iran using 27-year (1996-2022) data by employing two neural network models, i.e., stand-alone artificial neural network (ANN) and hybrid ANNHoney badger optimization algorithm (ANN-HBA). In temporal modeling, one-month and two-month precipitation delay steps were used as input parameters, and monthly precipitation was considered as output parameter. On the other hand, in spatial modeling, longitude, latitude, and altitude were used as input parameters and average monthly precipitation was considered as output parameter. Comparative performance of both the employed models was evaluated using three statistical criteria, i.e., root mean square error (RMSE), correlation coefficient (R) and Nash- Sutcliffe efficiency (NSE). In addition, precipitation in ungauged areas was estimated based on rainfall data of the nearby rain gauges through spatial interpolation methods such as isohyet and Thiessen polygon. Results of the temporal precipitation modeling revealed better performance of the hybrid ANN-HBA model over the stand-alone ANN model. Furthermore, the hybrid ANN-HBA model, during testing phase of the temporal modeling, could resemble the precipitation most accurately at Maraghe station (with values of R as 0.95, RMSE as 1 mm, and NSE as 0.85). Similarly, the hybrid ANN-HBA model during testing phase (with values of R as 0.95, RMSE as 1 mm, and NSE as 0.96) outperformed the stand-alone ANN model in depicting the precipitation adequately at spatial scale.

Optimal Design of Voltage Equalization Ring for the 1100 kV DC Voltage Proportional Standard Device Based on the Nation Standard Device Neural Network and Grey Wolf Optimization Algorithm

The DC voltage ratio standard device is an important tool for calibrating DC voltage transformers. At the 1100 kV voltage level, an increase in electric field intensity will increase the local heat generated inside the device, affecting the accuracy of its measurement. Using a suitable grading ring can even out the electric field intensity and reduce the maximum field strength to improve its measurement accuracy. This article mainly optimizes the design of the grading-ring structure of the 1100 kV DC voltage ratio standard device. First, a finite-element model of the 1100 kV DC voltage ratio standard device was built based on ANSYS; the electric field distribution around the voltage divider was calculated and analyzed, and a data set was constructed based on the calculation results. Secondly, for the optimization of electric field strength, this article presents the design of the nation standard device neural network, which learns the relationship between the structural parameters of the toroidal core and the field strength under the PyTorch 1.8 deep learning framework. Due to the strong convergence performance, few parameters, and ease of implementation of the grey wolf optimization algorithm, this study selected this algorithm to optimize the structural parameters of the grading ring. Finally, simulation examples are established in Python for validation. The experimental results indicate that the maximum field strength of the grading ring decreased from 12,161.1348 V/cm to 10,009.2881 V/cm, a reduction of 17.69%. The optimized structural parameters of the grading ring effectively reduced the electric field intensity around the 1100 kV DC voltage proportional standard device, providing a reliable and practical design approach for the selection of the DC voltage ratio standard device.

Non-volatile double-tunable vortex metalens design based on Sb2S3 using deep neural network and particle swarm optimization algorithm

A dual-tunable vortex metalens is proposed that leverages deep neural network (DNN) and particle swarm optimization algorithm (PSO), incorporating the use of the phase-change material Sb2S3, which enables simultaneous tunable of both the focal length of the focused vortex beam and the topological charge. The control of traditional vortex metalens relies on changing the polarisation state of the incident light; however, its control range is limited. We employed neural network algorithms to learn the complex mapping relationship between the unit structure and the phase, achieving a fast and accurate prediction of the vortex metalens's phase. Compared with traditional simulation methods, the application of this neural network significantly reduces the time consumption of the design process and avoids the need for re-simulation. Furthermore, our design transcends limitations associated with specific wavelengths and enables the rapid design of tunable metalens across the 1–1.5 μm wavelength range, offering flexibility for the development of dual-tunable metalens. Based on this scheme, we simulated vortex metalens with tunable focal length and topological charge, and the results showed that the simulation results were consistent with the theoretical predictions.

Intelligent optimization design of squeeze casting process parameters based on neural network and improved sparrow search algorithm

Squeeze casting process parameters are the key to squeeze casting production and to obtain excellent performance casts. To realize intelligent optimization design of process parameters under various requirements, this work presents a new intelligent optimization design framework for squeeze casting process parameters based on process data and integrating a two-stage intelligent integrated optimization. To adapt to diverse optimization applications of different process and target parameters, the backpropagation (BP) neural network and existing process data are utilized to intelligently establish the incompletely determined correlations between process parameters and squeeze cast quality or properties. An improved sparrow search algorithm (LCSSA) is developed and integrated to optimize both the aforementioned model structure and intelligently obtain the optimal solution, that is, the two stages of optimization. For the purpose of assessing the effect of each performance on the combination of process parameters, the information entropy weight approach is used. Various application cases and experiments have been conducted to evaluate the effectiveness of the suggested intelligent optimization design framework. It is indicated that the proposed method is practicable and can intelligently achieve ideal process parameters with high accuracy even based on small-scale data samples. The solving efficiency and optimization accuracy of LCSSA are superior than those of other intelligent optimization algorithms like the genetic algorithm (GA). The proposed framework outperforms the mixture of the BP neural network and other traditional optimization algorithms.

RETRACTED ARTICLE: Network security threat detection technology based on EPSO-BP algorithm

With the development of Internet technology, the large number of network nodes and dynamic structure makes network security detection more complex, which requires the use of a multi-layer feedforward neural network to build a security threat detection model to improve network security protection. Therefore, the entropy model is adopted to optimize the particle swarm algorithm to decode particles, and then the single-peak and multi-peak functions are used to test and compare the particle entropy and fitness values to optimize the weights and thresholds in the multi-layer feedforward neural network. Finally, Suspicious Network Event Recognition Dataset discovered by data mining is sampled and applied to the entropy model particle swarm optimization for training. The test results show that there are four functions for the optimal mean and standard deviation in this algorithm, with values of 5.712e − 02, 4.805e − 02, 4.914e − 01, 1.066e − 01, 1.577e − 01, 1.343e − 01, and 2.089e + 01, 5.926, respectively. Overall, the algorithm proposed in the study is the best. Finally, the detection rate of attack types is calculated. The multi-layer feedforward neural network algorithm is 83.80%, the particle swarm optimization neural network algorithm is 91.00%, and the entropy model particle swarm optimization algorithm is 95.00%. The experiment shows that the research model has high accuracy in detecting network security threats, which can provide technical support and theoretical assistance for network security protection.

Hybrid convolutional neural network and Flexible Dwarf Mongoose Optimization Algorithm for strong kidney stone diagnosis

This study presents a novel approach to kidney stone diagnosis using convolutional neural networks (CNN), applied to computed tomography (CT) images. The research addresses the challenge of data imbalance and protocol variation in medical imaging, which often leads to poor generalization of deep learning models. The model first uses three preprocessing techniques to enhance the quality of raw images and increase their quantity for effective CNN training. The main idea is to optimize the main arrangement of the convolutional neural network based on the proposed flexible version of dwarf mongoose optimization (FDMO) algorithm to provide a good detector model in kidney stone diagnosis. The model is then trained and tested on images from “CT Kidney Dataset”, and its comparison results with some other published works demonstrates its robustness and ability to generalize. The results indicate a significant improvement in diagnostic accuracy, potentially minimizing physician-induced errors and enhancing patient care.

Optimization of medium components for protein production by Escherichia coli with a high-throughput pipeline that uses a deep neural network

To optimize rapidly the medium for green fluorescent protein expression by Escherichia coli with an introduced plasmid, pRSET/emGFP, a single-cycle optimization pipeline was applied. The pipeline included a deep neural network (DNN) and mathematical optimization algorithms with simultaneous optimization of 18 medium components. To evaluate the DNN data sampling method, two methods, orthogonal array (OA) and Latin hypercube sampling (LHS), were used to design 64 initial media for each sampling method. The OA- and LHS-based data sampling resulted in green fluorescent protein fluorescence intensities of 0.088× 103-1.85× 104 and 3.30× 103-1.50× 104, respectively. Fifty DNN models were built using the OA and LHS datasets. Hold-out validation was performed using 15% test of OA and LHS data. Mean square errors of the DNN models were 0.015-0.64, indicating the estimation accuracies were sufficient. However, the sensitivities of components in the DNN models varied and were grouped into six major classes by the index of k-means clustering. A representative model was selected for each class. Mathematical optimization algorithms using Bayesian optimization and genetic algorithm were applied to the representative models, and representative optimized medium (OM) compositions were selected by k-means clustering from the proposed OMs. A total of 54 OMs were obtained from the OA and LHS datasets. In the validating cultivation, the best OMs of OA and LHS were 2.12-fold and 2.13-fold higher, respectively, than those of the learning data.

Multi-objective optimization design of aerodynamic and infrared characteristics of multi-stream serpentine nozzle

A multi-objective optimization design of multi-stream serpentine nozzle was carried out to improve the aerodynamic performance and reduce the infrared radiation signal. A numerical simulation was carried out to investigate the effects of nozzle pressure ratio (NPR), Aspect ratio (AR) and Ratio of length and diameter (L/D). Computational fluid dynamics (CFD) simulation was employed to study the aerodynamic performance of the multi-stream serpentine nozzle, and the narrow-band method with discrete transfer method were used to calculate its infrared radiation intensity distribution characteristics. A 3-factor, 21-level orthogonal table was designed based on the orthogonal experimental method, and multi-stream serpentine nozzle with different geometrical parameter and operating conditions were designed to obtain the aerodynamic performance and infrared radiation intensity data of each experimental sample point. Radial Basis Function (RBF) neural network and Multi-Objective Particle Swarm Optimization algorithm were used to optimize the geometry and operating conditions with better aerodynamic performance and weaker infrared signal. The result shows that the Pareto solutions obtained by the optimization have good accuracy compared with the numerical simulation results, with an aerodynamic performance error of 0.02% and an infrared radiation intensity error of 0.24%.

Neural network-based modeling of solid oxide fuel cells for marine applications

As the solid oxide fuel cell (SOFC) experimental test is still quite cost-effective and time-consuming, there is a growing need for developing effective simulation tools to reduce the time and cost of the marine SOFC performance test and optimization. The present paper is aimed to study the modeling and simulation of marine solid oxide fuel cells by artificial intelligence method. A neural network based on a particle swarm optimization algorithm is used to establish a marine solid oxide fuel cell model for voltage/current characteristic analysis. The model is also compared with BP neural network and Hopfield neural network methods. The simulation results compared with experimental data show that the effectivity of the particle swarm optimization neural network algorithm is best, which can accurately predict the voltage/current characteristic curves of a SOFC under different fuel flow-air volume ratios. The model study can provide support for SOFC performance characteristics analyses and has significant potential in SOFC optimization applications.

Electric vehicle charging design: The factored action based reinforcement learning approach

Charging optimization design for Electric Vehicles (EV) is challenging because it must account for various uncertainties and design aspects such as random EVs’ arrivals and departures, battery degradation, and transformer Loss of Life (LoL). Model-free reinforcement learning (RL) can be employed to tackle such the EV charging design where it does not require to explicitly model the environment dynamics and accurately predict relevant system parameters. However, the high complexity involved in conventional RL-based approaches usually limits its application to only small-scale EV charging settings, which is impractical. To overcome this limitation, we employ the factored action based RL method to transform the formulated Markov Decision Process (MDP). Then, we propose novel reward shaping and hybrid learning methods combining the Convolutional Neural Network (CNN) and Proximal Policy Optimization (PPO) algorithm to extract relevant features from high-dimension state space and efficiently solve the transformed MDP problem. Extensive numerical studies demonstrate that the proposed design can be used to control a charging station (CS) supporting a large number of EVs. Moreover, we show that the proposed framework greatly outperforms other baselines including single-agent and multi-agent RL based strategies and a heuristic power scheduling algorithm in terms of the achieved reward.

Radial basis function neural network optimization algorithm based on dynamic inertial weight particle swarm optimization for separating overlapping peaks in ion mobility spectrometry.

Ion mobility spectrometry (IMS), as a promising analytical tool, has been widely employed in the structural characterization of biomolecules. Nevertheless, the inherent limitation in the structural resolution of IMS frequently results in peak overlap during the analysis of isomers exhibiting comparable structures. The radial basis function (RBF) neural network optimization algorithm based on dynamic inertial weight particle swarm optimization (DIWPSO) was proposed for separating overlapping peaks in IMS. The RBF network structure and parameters were optimized using the DIWPSO algorithm. By extensively training using a large dataset, an adaptive model was developed to effectively separate overlapping peaks in IMS data. This approach successfully overcomes issues related to local optima, ensuring efficient and precise separation of overlapping peaks. The method's performance was evaluated using experimental validation and analysis of overlapping peaks in the IMS spectra of two sets of isomers: 3'/6'-sialyllactose; fructose-6-phosphate, glucose-1-phosphate, and glucose-6-phosphate. A comparative analysis was conducted using other algorithms, including the sparrow search algorithm, DIWPSO algorithm, and multi-objective dynamic teaching-learning-based optimization algorithm. The comparison results show that the DIWPSO-RBF algorithm achieved remarkably low maximum relative errors of only 0.42%, 0.092%, and 0.41% for ion height, mobility, and half peak width, respectively. These error rates are significantly lower than those obtained using the other three algorithms. The experimental results convincingly demonstrate that this method can adaptively, rapidly, and accurately separate overlapping peaks of multiple components, improving the structural resolution of IMS.

Prediction of Transformer Oil Temperature Based on an Improved PSO Neural Network Algorithm

Abstract: In addressing the issue of power transformer oil temperature prediction, traditional back propagation (BP) neural network algorithms have been found to suffer from local optimization and slow convergence. This study proposes an oil temperature prediction model based on an improved particle swarm optimization (PSO) neural network algorithm, which introduces an asymmetric adjustment learning factor and a mutation operator. The BP neural network, genetic algorithm (GA) optimization neural network, and the improved PSO neural network are compared by considering various factors, such as ambient temperature, load changes, and the number of cooler groups under different working conditions. Results show that the proposed algorithm improves the actual change trend of oil surface temperature and makes the transformer operation more stable to a certain extent. Background: The mathematical model for predicting transformer oil temperature is clear, but the parameters in the model are uncertain and vary with time. When subjected to different operating conditions, such as ambient temperature, load changes, and the number of cooler groups acting independently or in combination, the prediction results of the oil temperature model vary with different system parameters. Objective: This paper aims to enhance the accuracy of transformer temperature prediction. In order to optimize the oil temperature prediction model, asymmetric adjustment learning factors and mutant operators are added to meet diverse system parameter requirements. Methods: The paper utilizes an oil temperature prediction model based on an improved PSO neural network algorithm, which introduces an asymmetric adjustment learning factor and a mutation operator to address the limitations of the standard PSO algorithm. Results: This paper has employed a fusion algorithm of the genetic algorithm of the BP neural network and the PSO algorithm, and conducted simulation and experimental analysis. The simulation and experimental results demonstrate the accuracy and effectiveness of the fusion algorithm. Conclusion: This study demonstrates enhanced prediction accuracy of transformer oil surface temperature using the improved particle swarm optimization neural network algorithm. This algorithm has less prediction error under different working conditions compared to other algorithms. By increasing population diversity and combining inertia weights, the algorithm not only greatly improves its search performance but also avoids local optimization.

Thermal management of methanol steam reforming reactor by back propagation neural network and particle swarm optimization algorithm

Methanol steam reforming (MSR) reactor is a device that converts the methanol into hydrogen. Since the MSR is an endothermic reaction, the methanol conversion is usually affected by the heating conditions. So, to improve the methanol conversion, the heating condition should be managed. This study aims to present a framework and methodology for studying the thermal management of a determined MSR reactor by controlling the heating conditions. Firstly, the characteristics of a three-dimensional MSR reactor are carried out, and the effects of the inlet condition parameters e.g. heating parameters and fuel supply parameters, on the methanol conversion are investigated. Then, the performance prediction of the MSR reactor is studied using several common intelligence algorithms and the corresponding surrogate models are obtained. The back propagation neural network (BPNN) surrogate model shows the best prediction accuracy than stepwise linear (SL), radial basis function neural network (RBFNN), linear support vector machines (L-SVM), and random forest algorithms (RFA). Finally, the thermal management of the MSR reactor is carried out by combining the surrogate model and particle swarm optimization (PSO) algorithm to obtain better performance. The results show that hydrogen production of the MSR reactor can be generally guaranteed by the characteristic curves of the heating parameters and fuel supply parameters, and the methanol conversion can be maintained steadily above 93% within the whole period of the hydrogen production requirements.

Design of Bismuth-Doped Fiber Amplifier Based on Neural Network and Multi-Objective Optimization Algorithm

Design of Bismuth-Doped Fiber Amplifier Based on Neural Network and Multi-Objective Optimization Algorithm

A long-term prediction method for PM2.5 concentration based on spatiotemporal graph attention recurrent neural network and grey wolf optimization algorithm

With the rapid advancement of global industrialization and urbanization, the problem of air pollution, primarily attributed to PM2.5, as the main component has become increasingly serious, directly affecting people's health. Therefore, it is necessary to forecast the long-term PM2.5 levels. However, most existing prediction methods for PM2.5 concentration lack sufficient extraction of urban spatial characteristics. Therefore, this paper proposes a novel method based on spatiotemporal graph attention recurrent neural network and grey wolf optimization (GWO-GART). Firstly, a long-term prediction model of PM2.5, the Graph Attention Cycle Network (GART), is constructed by using Graph Attention network (GAT), Graph Neural Network (GNN) and Rated Recurrent Unit (GRU) networks. In the GART model, a novel multi-GAT is proposed for the first time, which can extract the spatial features of cities circularly and obtain more implicit associations between cities. Secondly, we utilize grey wolf optimization (GWO) to select the best set of hyperparameters for the model. This automated approach aids in designing the network structure of GART. We compared GWO-GART with three single models and two hybrid models using a dataset collected from the real world. The results show that, compared with the state-of-the-art (SOTA) models, GWO-GART had an average reduction of 2.13% in RMSE, 2.47% in MAE, and an average increase of 1.55% in CSI on all datasets. This demonstrates that GWO-GART provides an efficient approach to precisely forecast long-term PM2.5 levels in urban regions.