International Journal of Modelling, Identification and Control (16 papers in press)

Regular Issues

• Open Access

  Improved BP algorithm for measurement of urban business environment based on machine learning ( Free Full-text Access )

  by Huihui Deng, Weitao Liu, Zhuolin Xiao, Yaxi Zhang 
  Abstract: To overcome these limitations, this study proposes a machine learning-driven urban business environment measurement model based on an improved Back Propagation Neural Network (BPNN). First, the core constituent dimensions of the business environment are systematically reviewed. The proposed model is compared against benchmark models including the standard BP, Support Vector Machine (SVM), and Random Forest (RF). Performance evaluation results demonstrate that the proposed improved BP model achieves significantly superior performance across key metrics including Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and coefficient of determination (R-squared, R2), enabling more accurate fitting and prediction of urban business environment levels. This study provides an efficient and objective new paradigm for the quantitative assessment of business environments and offers valuable references for intelligent measurement studies of other complex socio-economic systems.
  Keywords: business environment; BP neural network; machine learning; improved algorithm; double optimisation mechanism; support vector machine; SVM.
  DOI: 10.1504/IJMIC.2026.10080695
   
  

• Open Access

  Modelling and identification of intelligent teaching system for international trade English under the Belt and Road Initiative ( Free Full-text Access )

  by Yaoyuan Zhang 
  Abstract: Against the Belt and Road Initiative, traditional international trade English teaching lacks quantitative student ability identification and personalized learning scheduling. This paper establishes a complete data-driven intelligent teaching modelling and identification framework. Firstly, K-means clustering is adopted to identify students multi-dimensional language proficiency grades from listening, speaking, reading and writing scores. Apriori association rule mining extracts internal correlations among language modules to locate learning weak points. A genetic algorithm is constructed for offline learning resource allocation, and deep reinforcement learning (Actor-Critic) is integrated for dynamic personalized learning path optimization. Meanwhile, Transformer NLP model is embedded to build automated writing and oral evaluation subsystems for language feature identification. Experiments on 198 undergraduates verify that the proposed hybrid model achieves significantly higher score improvement (average 19.2% across four modules) than traditional single-algorithm schemes. The proposed modelling and identification method provides a quantifiable control strategy for customised trade English teaching.
  Keywords: modelling and identification; K-means clustering; association rule mining; deep reinforcement learning; trade English intelligent teaching.
  
  

• Open Access

  Modelling and identification of the mediating mechanism of self-efficacy between creative thinking and academic burnout under AI educational context ( Free Full-text Access )

  by Na Yu, Fanghui Lou 
  Abstract: The rapid advancement of Artificial Intelligence (AI) technology is profoundly reshaping the teaching models and future employment landscape of higher education. It poses unprecedented challenges to college students' cognitive patterns and psychological adaptability. Against this backdrop of transformation, academic burnout among college students has become increasingly prevalent. However, research exploring its mitigation mechanisms from the perspective of positive psychological resources remains insufficient. This study examines the relationships among creative thinking, self-efficacy, and college student academic burnout in the AI era, along with their underlying mechanisms. Through theoretical construction and questionnaire surveys, data analysis was conducted on 168 valid samples of college students. This should be regarded as an important pathway for enhancing students' psychological resilience and strengthening their sense of academic achievement. This will effectively address the educational challenges brought by the AI era.
  Keywords: creative thinking; ideological and political teaching; educational psychology; learning burnout; mental health; artificial intelligence.
  DOI: 10.1504/IJMIC.2026.10081108
   
  

• Open Access

  A multivariable identification model for entrepreneurial performance control in higher education under AI technology ( Free Full-text Access )

  by Yanling Chen, Jiaxuan Wei 
  Abstract: The entrepreneurial performance of university teachers and students is influenced by a complex interplay of psychological traits, ideological education, and technological capabilities, yet systematic quantitative modeling of these factors remains limited. This paper constructs a multivariable linear regression system to identify the core drivers of entrepreneurial performance in the context of AI-driven innovation. Two questionnaire-based datasets were collected from university teachers and students, yielding 255 and 270 valid samples, respectively. Pearson correlation analysis and hierarchical linear regression modeling were employed to estimate the direct effects of psychological capital, ideological education, and AI application ability on entrepreneurial performance. These findings provide a measurable quantitative framework for optimizing university entrepreneurship education systems and implementing targeted cultivation strategies. The proposed modeling approach offers a systematic method for identifying and quantifying the key parameters influencing entrepreneurial outcomes in educational contexts.
  Keywords: variable identification; entrepreneurial performance; psychological capital; ideological education; artificial intelligence literacy.
  DOI: 10.1504/IJMIC.2026.10081282
   
  

• Open Access

  Machine learning models and anomaly detection algorithms for audit data consistency verification ( Free Full-text Access )

  by Piaopiao Tan, Yongjiao Yang, Jiaxin Lin 
  Abstract: Under the background of digital transformation of enterprises, massive heterogeneous financial data brings verification challenges. Traditional rule matching and manual sampling models have limitations in efficiency and coverage. Therefore, a hybrid strategy model combining supervised classification and unsupervised anomaly detection is constructed to deal with the dual characteristics of clear known violation patterns but hidden unknown anomalies. Its purpose is to realize the automatic verification of audit data consistency and high-precision anomaly identification. Starting with the deep mining of business logic dependence, this article designs MLP (Multilayer Perceptron Machine) for feature mapping, and combines SVM (Support Vector Machine) with IForest (Isolated Forest) score enhanced by self-encoder reconstruction error, and balances known risk location and unknown outlier discovery through dynamic weight mechanism.
  Keywords: anomaly detection; data consistency; machine learning; feature engineering.
  DOI: 10.1504/IJMIC.2026.10081461
   
  

• Open Access

  OOptimisation of teaching effectiveness of ideological and political education by integrating artificial intelligence and graph convolutional network ( Free Full-text Access )

  by Xu Gao, Xiaolong Liang 
  Abstract: In the context of deepening education informatisation in the new era, ideological and political education (IPE) is a core link to fulfil the fundamental task of fostering virtue and integrity; it confronts multiple challenges (such as diversified student ideologies and the growing demand) for personalised learning. The conventional one-size-fits-all teaching mode fails to achieve the goal of targeted talent cultivation; it creates an urgent need for intelligent technical tools to boost teaching quality. This study proposes an optimisation method for teaching effectiveness of IPE based on GCN. GCN is used to extract student features and predict teaching performance by constructing a student learning relation graph. On the whole, this study provides a new technical path and theoretical support for the intelligent transformation of IPE; it also has an important practical value for promoting the construction of the all-staff, all-process, and all-round education system.
  Keywords: graph convolutional network; GCN; ideological and political education; IPE; artificial intelligence; teaching optimisation; graph neural network; GNN; personalised teaching.
  DOI: 10.1504/IJMIC.2026.10081462
   
  

• Open Access

  A multimodal modelling and recognition method for audio visual aesthetic teaching in college vocal music using deep learning ( Free Full-text Access )

  by Yanan Li 
  Abstract: The integration of multimedia artificial intelligence into college vocal music education presents a challenging modelling problem: how to jointly represent audio signals (pitch, rhythm, timbre) and visual information (facial expression, mouth movement, body gesture) for aesthetic perception and teaching feedback. This study proposes a modelling and recognition framework for audio-visual aesthetic teaching in university vocal music courses under the STEAM educational philosophy. The framework employs an encoder-decoder architecture with an attention mechanism to achieve joint representation of audio and visual modalities. The audio feature extraction module uses Mel-frequency cepstral coefficients combined with a convolutional neural network, while the visual feature extraction module processes video frame sequences using 3D convolution and ResNet. Subjective evaluation from 137 university students indicates positive feedback on learning interest, aesthetic understanding, and classroom engagement. This modelling approach provides a systematic framework for intelligent vocal music education that links sensory perception, aesthetic experience, and feedback control.
  Keywords: audio-visual modelling; multimodal recognition; encoder-decoder architecture; vocal music education; deep learning.
  DOI: 10.1504/IJMIC.2026.10081570
   
  

• Open Access

  Students behaviour modelling and learning progress prediction algorithm in intelligent education evaluation system ( Free Full-text Access )

  by Congcong Ge 
  Abstract: As online education shifts from outcome-based assessment toward continuous, process-oriented evaluation, accurately identifying students' behavioral patterns and forecasting their learning progress has become a central challenge. This study proposes an intelligent educational evaluation framework that integrates multi-source behavioral features. Heterogeneous eventsincluding clickstream records, responses, and video interactionsare encoded in a unified representation to construct a dynamic student-state sequence. A Transformer encoder with causal self-attention captures long-range dependencies and produces a comprehensive cognitive-state vector. A dual-task prediction head jointly estimates next-response correctness and medium- to long-term knowledge mastery, thereby supporting multi-step learning-progress prediction. Experiments on the ASSISTments and Junyi datasets show that the proposed method outperforms the evaluated baselines: AUC increases by 6.8%9.6%, RMSE decreases by 20.9%, and the ablation results confirm the importance of multi-source behavior, positional encoding, and causal masking.
  Keywords: intelligent education evaluation; knowledge tracking; student behaviour; prediction of learning progress; transformer.
  DOI: 10.1504/IJMIC.2026.10081757
   
  

• Open Access

  BE-SCNet: multi-modal feature identification model integrating LSTM, BERT and graph convolution for student academic forecasting ( Free Full-text Access )

  by Yanjun Bu, Min Chen 
  Abstract: Traditional academic performance prediction models primarily rely on single-source structured data and often fail to capture students' emotional states and social interaction characteristics, resulting in limited prediction robustness and stability. This paper proposes a multimodal deep learning framework, termed BE-SCNet, for academic success prediction. The proposed framework integrates three complementary subnetworks: a Long Short-Term Memory (LSTM) network for learning behavior sequence modeling, Bidirectional Encoder Representations from Transformers (BERT) for textual sentiment feature extraction, and a Graph Convolutional Network (GCN) for social network topology modeling. Furthermore, an adaptive weighted fusion mechanism is introduced to dynamically balance the contributions of heterogeneous multimodal features. Experiments are conducted on real-world multi-source university datasets to evaluate the model's effectiveness and generalization capability. Overall, the proposed data-driven multimodal learning framework provides an effective and reliable predictive decision-support tool for intelligent higher education management systems.
  Keywords: system modelling; feature identification; multimodal deep learning; graph convolution; academic performance prediction; predictive control; graph convolutional network; GCN.
  DOI: 10.1504/IJMIC.2026.10081758
   
  

• Open Access

  Multi-modal feature fusion recognition of face and gait under partial clothing occlusion in autumn and winter ( Free Full-text Access )

  by Liuren Wang 
  Abstract: Aiming at severe accuracy degradation of single-modal biometric recognition under autumn and winter clothing occlusion in cold-region outdoor surveillance, this paper proposes a lightweight occlusion-aware dynamic weight fusion algorithm for face and gait biometrics. A dual-branch network extracts facial and skeleton-based gait features. An occlusion quantification module calculates feature confidence based on pixel-level occlusion ratios and skeleton keypoint integrity. Softmax-normalized dynamic weights adaptively balance the two modalities, while depthwise separable convolution and knowledge distillation jointly realize lightweight optimization. Experiments conducted on public datasets and a self-built paired occlusion dataset verify that the proposed method outperforms fixed-weight and single-modal algorithms by 8.7 to 15.2% in recognition accuracy, with a backbone inference speed of 32 FPS that satisfies national real-time monitoring standards.
  Keywords: multi-modal biometric recognition; clothing occlusion; dynamic weight fusion; lightweight network; occlusion quantification; face-gait fusion; edge surveillance.
  DOI: 10.1504/IJMIC.2026.10081759
   
  

• Open Access

  The application of artificial intelligence and deep learning models in ideological and political education at colleges and universities ( Free Full-text Access )

  by Xiaoqing He, Weijun Liu 
  Abstract: With the advancement of information technology, traditional methods of ideological and political education face new challenges and opportunities. This work explores the potential application of deep learning models in ideological and political education at the undergraduate level. By integrating natural language processing and sentiment analysis technologies, a novel deep learning framework is proposed to enhance teaching effectiveness and student engagement. Methodologically, a hybrid model combining the Convolutional Neural Network and Recurrent Neural Network is constructed. Moreover, its performance is ensured through hyperparameter optimization. Experiments based on over 100,000 pieces of student feedback data collected from multiple universities validate the model. Results show an accuracy of 85.2% and an F1 score of 84.4%, significantly improving students knowledge retention and interest levels. These findings confirm that artificial intelligence technologies can effectively promote the personalization and intelligence of ideological and political education, and offer new pathways for future educational innovation.
  Keywords: deep learning; ideological and political education; undergraduate education; artificial intelligence.
  DOI: 10.1504/IJMIC.2026.10081760
   
  

• Open Access

  Temporal attribution and feature mining of Erhu music emotion based on explainable deep learning ( Free Full-text Access )

  by Bin Li 
  Abstract: Music emotion recognition is an important research direction in the fields of intelligent music analysis and multimedia information understanding, with significant implications for understanding musical content, assisting music composition, and advancing intelligent music interaction. However, existing methods still face several challenges in complex musical environments, including insufficient extraction of local fine-grained features, difficulties in modeling long-term emotional dependencies, and a lack of interpretability in the model decision-making process. Erhu music emotion recognition involves complex temporal variations and acoustic feature interactions. Traditional deep learning models still exhibit limitations in local detail extraction, long-term dependency modeling, and explainable emotion decision-making. To address these issues, this study proposes an Explainable Parallel Local Attention Transformer (EPLAT) for temporal attribution and feature mining of Erhu music emotion. The model integrates parallel local attention structures with multi-layer Transformers. This study provides methodological references for Erhu music emotion analysis and explainable music intelligence studies.
  Keywords: music emotion recognition; explainable deep learning; transformer; temporal attribution; acoustic feature mining.
  DOI: 10.1504/IJMIC.2026.10081778
   
  

• Open Access

  Research on intelligent grounding wire status monitoring system based on IoT sensor network ( Free Full-text Access )

  by Yuxin Zhu, Rong Chang, Yanbin Fang, Meng Qin, Fushun Cha 
  Abstract: In power system maintenance operations, the missing or mistaken removal of grounding wires can lead to severe casualty accidents. To address this critical industrial safety pain point, this paper proposes an intelligent grounding wire status monitoring system based on an Internet of Things (IoT) sensor network. The integrated system features three core technical innovations: a multi-modal fusion sensing architecture covering electrical, mechanical, and thermal parameters; an adaptive LoRa communication protocol dynamically optimizing parameter configurations via a composite Channel Quality Indicator (CQI); and a lightweight edge intelligent recognition model, GW-Net. A comprehensive 6-month on-site validation at a 220kV substation indicates that the proposed GW-Net model achieves a high F1 score of 0.971 with only a 0.4MB parameter size. Moreover, the adaptive communication scheme ensures a system availability of 99.95% under electromagnetic interference, and the 99th percentile intra-station edge closed-loop response latency is restricted to 310ms, strictly satisfying the 500ms safety limit and achieving zero missed alarms in practical field tests.
  Keywords: intelligent grounding wire monitoring; internet of things; IoT sensor network; multi-sensor fusion; edge computing; lightweight deep learning; power safety; power anti-misoperation; edge interlock; channel quality indicator; CQI.
  DOI: 10.1504/IJMIC.2026.10081783
   
  

• Open Access

  Collaborative innovation of visual narrative for rural culture of Bailuyuan and college student education empowered by AIGC technology ( Free Full-text Access )

  by Linlin Chen, Jinxia Wang 
  Abstract: The visual narrative of rural culture in Bailuyuan faces several problems in the process of integrating into college education. The application of digital preservation technology is superficial, which fails to dig deep into cultural connotation; Lack of narrative forms that fit the aesthetics and cognition of young audiences; There is a lack of systematic transformation path between cultural resources and educational practice. The purpose of this study is to construct the cultural knowledge map and multi-modal generation model of Bailuyuan through Artificial Intelligence Generated Content (AIGC) technology, and realize the intelligent extraction, reconstruction and dissemination of cultural elements. The study builds upon the rich rural cultural resources of the Bailuyuan region. It constructs an AIGC-based framework for rural culture visual narrative. This study enriches the application of AIGC in cultural heritage protection and offers valuable insights for talent cultivation models under the rural revitalization strategy.
  Keywords: artificial intelligence generated content; AIGC; Bailuyuan; rural culture; visual narration; college education; collaborative innovation.
  DOI: 10.1504/IJMIC.2026.10082019
   
  
• Outlier detection algorithm based on deviation characteristic  
  by Yong Wang, Hongbin Wang, Pengcheng Sun, Xinliang Yin 
  Abstract: Outlier mining focuses on researching rare events through detection and analysis to dig out the valuable knowledge from them. In the static data set environment, the traditional LOF algorithm calculates the local outlier factor through the whole data set and requires a lot of computing time. To solve this problem, the algorithm divides the data space into grids, and calculates the local outlier factor based on the centroids of the grids. Since the grid number is less than data point number, the time complexity is obviously reduced under acceptable error. When the new data points are added, it can rapidly detect outliers. The contrast experiment results show that the new algorithm can reduce the computation time and improve the efficiency, while achieving comparable accuracy.
  Keywords: outlier detection; local outlier factor; deviation characteristic; fast LOF detection algorithm.
  
  
• A novel instance segmentation method for lane instance detection  
  by Tengyu Guo, Yuexin Hao, Xiaolong Mu 
  Abstract: Over the past decades, lane detection has attracted considerable research attention, with prior efforts predominantly targeting single-lane or adjacent lane scenarios. However, advancing demands for vehicle automation necessitate precise lane-level localisation, requiring the detection of an arbitrary number of lanes within complex environments. To address this challenge, we propose a novel instance segmentation framework that robustly identifies individual lane instances. Unlike conventional approaches, our method introduces a similarity module that treats segmentation as a clustering task. This module calculates pixel-to-pixel similarity scores throughout the image, generating per-pixel similarity masks. These masks are then aggregated into a condensed set of lane predictions using an efficient, parallelisable clustering algorithm. Furthermore, we integrate our similarity module into a semantic segmentation network (e.g., DeeplabV3+ or ENet) by branching from the networks mid-level features, thereby leveraging its hierarchical semantic understanding while enhancing instance-aware segmentation through our similarity-driven clustering. Experimental results demonstrate the frameworks effectiveness, achieving lane detection accuracies of 97.1% and 96.2% with DeeplabV3+ and ENet backbones, respectively. These results underscore the robustness and precision in multi-lane scenarios, offering a promising foundation for advancing lane-level localisation in autonomous driving systems.
  Keywords: road lane detection; cluster; instance segmentation; deep neural network.
  DOI: 10.1504/IJMIC.2026.10080919