Song Qingqing - Mathematical Modeling & Industrial Software Systems on Song Qingqing

A Database Load Analysis Method for Query-Intensive Data Storage Processing Systems

Mon, 12 Apr 2021 12:00:00 +0000

Scientific Seminar on Mathematical Modeling, Data Analysis, and Information Security of Complex Systems — Invited Talk

Fri, 17 Apr 2026 12:00:00 +0000

In 2026, delivered an invited talk entitled ‘Performance Time Series Analysis in the Multiple Baseline Framework’ at the BSU APMI Scientific Seminar. The talk articulated that PTS analysis research should focus on three core components: (1) multi-baseline modeling and switch detection; (2) multi-baseline-aware anomaly detection mechanisms; and (3) interpretable continuous health scoring for system state quantification.

OS-Metrics: A Performance Time Series (PTS) Dataset for Anomaly Detection and Change Point Detection

Wed, 14 Jan 2026 00:00:00 +0000

OS-Metrics is a 7-day, 15-second-resolution, causally-annotated performance time series dataset for evaluating anomaly detection and change point detection algorithms. It contains 6 metric files and 3 event log files.

OS-Metrics is evaluation-ready: all 42 perturbation events are causally annotated, including start/end time, intensity, mode, and affected metrics.
Metrics are continuously sampled at 15-second intervals over 7 days.
Event logs use zero-based indexing (start_point, end_point), aligned with the metric time series (first row index is 0).
Remote observation bias (network_response_time_diff.csv) reflects genuine network path asymmetry, not artificial error or artifact.
All timestamps are in UTC; no daylight saving time adjustments have been applied.

Details are provided in the README.md file. This dataset is also available on Zenodo: https://doi.org/10.5281/zenodo.17981352 (the Mendeley version mirrors the Zenodo version).

Two Invited Talks at the XII Belarusian-Chinese Youth Innovation Forum "New Horizons — 2025"

Thu, 27 Nov 2025 12:00:00 +0000

In 2025, at the XII Belarusian-Chinese Youth Innovation Forum ‘New Horizons — 2025’ held in Minsk, delivered two invited talks on the foundational methodology of Performance Time Series (PTS).

Brain Cancer MRI Classification Using Convolutional Neural Network Architectures Optimized via Grid Search and Coordinate Ascent

Mon, 29 Sep 2025 00:00:00 +0000

Magnetic Resonance Imaging (MRI) is an important tool for brain cancer diagnosis and classification. Combined with modern Convolutional Neural Network (CNN) technology, it can effectively improve the accuracy and efficiency of tumor classification, providing critical references for clinicians. Prior studies have demonstrated the advantages of CNNs in medical image classification. However, in-depth investigations into the performance differences among CNN models for brain cancer MRI image classification tasks and their hyperparameter optimization remain insufficient, limiting model selection and accuracy improvement in clinical applications. This study aims to provide a CNN model selection and optimization methodology for brain cancer MRI image classification to improve classification accuracy and reliability. CNN models of varying depth and complexity, including LeNet-5, AlexNet, and ResNet-18, are selected, and Grid Search Enhanced with Coordinate Ascent (GSECA) is employed for hyperparameter optimization, furnishing a feasible model selection and optimization methodology for brain cancer MRI image classification. Experimental results demonstrate that the proposed method achieves excellent performance on the PMRAM four-class brain cancer MRI image dataset from Bangladesh. The ResNet-18-based classifier attains a test accuracy of 95.69%, with F1 scores of 94.33% (Glioma), 94.07% (Meningioma), 96.12% (Normal), and 98.21% (Pituitary) across the four classes, respectively. Notably, the precision for the Pituitary class reaches 99.10%. These results significantly outperform AlexNet and LeNet-5, ranking at an excellent level among comparable studies and achieving effective model selection. Meanwhile, the importance of network depth and effective hyperparameter optimization for boosting classification performance in such tasks is confirmed.

Change-Point Detection Utilizing Normalized Entropy as a Fundamental Metric

Thu, 18 Sep 2025 00:00:00 +0000

This paper proposes a change-point detection concept based on normalized entropy as a fundamental metric, aiming to overcome the limitations of traditional entropy-based methods that depend on distributional assumptions and absolute scale. Normalized entropy maps entropy values to the [0,1] interval through standardization, thereby accurately capturing relative changes in data complexity. The proposed method employs a sliding window to compute normalized entropy, transforming the problem of detecting change points in complex time series — induced by changes in scale, distribution, and diversity — into the problem of identifying salient features within the normalized entropy sequence. This circumvents the interference of parametric assumptions and effectively highlights distributional shifts. Experimental results demonstrate that normalized entropy exhibits pronounced numerical fluctuation characteristics and patterns near change points under various distributions and parameter combinations; the mean deviation between fluctuation moments and actual change points is only 2.4% of the sliding window size, indicating strong adaptability. This work provides theoretical support for change-point detection in complex data environments and lays a methodological foundation for precise, automated change-point detection using normalized entropy as a base metric.

TRLLD: Load Level Detection Algorithm Based on Threshold Recognition for Load Time Series

Wed, 16 Apr 2025 00:00:00 +0000

Load time series analysis is critical for resource management and optimization decision-making, particularly automated analysis techniques. Existing research insufficiently interprets the holistic characteristics of samples, resulting in markedly divergent load level detection conclusions for samples with different characteristics (trend, seasonality, periodicity). Achieving automated, feature-adaptive, and quantifiable analysis remains a challenge. This paper proposes a Threshold Recognition-based Load Level Detection algorithm (TRLLD) that, based on sample characteristics, effectively identifies regions of different load levels in samples of arbitrary scale and distribution type. The algorithm classifies data points using distribution density uniformity, ultimately yielding normalized load values. In the feature identification step, the algorithm employs a Difference-based Uniformity Index of Density (DUID), High Load Level Concentration (HLLC), and Low Load Level Concentration (LLLC) to assess sample characteristics; these metrics are independent of specific load values, thereby providing a standardized feature perspective and ensuring high efficiency and strong interpretability. Compared with traditional approaches, the proposed method demonstrates superior adaptivity and real-time analysis capability. Experimental results show that the method can effectively identify high-load and low-load regions across 16 time series samples with diverse load characteristics and produce highly interpretable results. The correlation between DUID and sample density distribution uniformity reaches 98.08%. When noise at 10% MAD intensity is introduced, the maximum relative error is 4.72%, demonstrating high robustness. Moreover, the method exhibits notable advantages in both general-sample and low-sample scenarios.

Trend-Aligned Feature Correlation: A Holistic Feature Relevance Metric on Time Series Data

Thu, 20 Mar 2025 00:00:00 +0000

Existing time series correlation research methods tend to over-rely on global statistical features while neglecting the dynamic variation of data along the temporal dimension, particularly falling short in trend inflection point identification and complex correlation discrimination. This study proposes an innovative Trend-Aligned Feature Correlation (TAFC) matching method that overcomes the limitations of traditional correlation analysis. TAFC employs a refined data preprocessing pipeline and introduces two unique metrics: the Trend Overlap Index (Q) and the Numerical Correlation Metric (P). The P metric effectively captures numerical correlation between data by combining a Normalized Difference Metric (NDM) with the Pearson correlation coefficient. Furthermore, TAFC introduces a customized granularity smoothing mechanism to balance trend feature preservation with noise suppression, aiming to comprehensively assess the multidimensional, multi-granularity correlation of time series data. This study constructs simulated scenario data based on chi-square and multimodal distributions for experimental validation, covering 36 sets of simulated data. Results demonstrate that the method can significantly differentiate datasets with distinct trend characteristics; the introduction of granularity has been validated to help make trend features more prominent; simultaneously, under the introduction of 5% random noise, the algorithmic error is controlled within the range of 5.60%–12.24%. This work provides a novel and precise approach for computer-aided time series data analysis with significant application value.

Utilizing Pretrained Convolutional Neural Networks for Classification of Lorenz Plots of RR Intervals

Sat, 01 Feb 2025 00:00:00 +0000

Background: Heart rate variability (HRV) and RR-interval Lorenz plots are key indicators for assessing cardiac health status. Lorenz plots can effectively characterize the dynamic changes of heart rate, yet their classification presents challenges. Currently, mathematical computation serves as the primary classification approach, but issues such as difficulty in region delineation and pattern complexity persist. Convolutional neural network models have proven effective across numerous problems; however, their application to Lorenz plot classification remains relatively limited.

Architecture Design of Highly Customized Document Generation Services in SaaS Systems Based on JACOB

Tue, 03 Dec 2024 00:00:00 +0000

Documents serve critical functions including information storage, work-efficiency enhancement, and collaborative innovation, playing an important role across various domains. In SaaS environments, efficient and flexible document generation services constitute not only the core of business processes but also a key instrument for enhancing competitiveness and driving business growth. Highly customized document generation services can effectively handle documents with high information density and diverse data representations, thereby improving data readability, operability, and flexibility. This paper completes the architectural design of highly customized document generation services in SaaS mode based on JACOB and the Spring Cloud microservices framework, achieving independent service scaling and elastic scaling, as well as data isolation and security guarantees. When generating identical content, the time difference between single-tenant document generation and multi-tenant concurrent document generation ranges from 0.69% to 4.7%, exhibiting nonlinear growth, though the performance degradation differential is not significant. When generating documents containing multiple images, document generation speed declines markedly, whereas for documents with fewer images, the character generation speed increases substantially.

Node Load Rate Calculation Method Based on Database Query Statistics in Load Balancing Strategy

Wed, 02 Oct 2024 00:00:00 +0000

The calculation of node load rate critically impacts the quality of load balancing strategies in distributed systems. Existing commonly used node load rate calculation methods fail to adequately account for the current actual state of nodes, resulting in discrepancies between computed values and actual conditions. This paper focuses on distributed information systems, taking the database as an entry point, and proposes a node load rate calculation method based on database query statistics suitable for load balancing strategies. This method employs an analytical approach combining theoretical computation data with benchmark performance test data, comprehensively accounting for factors such as data volume in the distributed information system database, peak database access periods, and data table indices that affect computation results. Under the premise of ensuring the accuracy of node load rate computation, the method minimizes repetitive testing, conserves computational performance and energy resources, and reduces both the time and economic costs of node load rate calculation. Algorithm validation results show that the mean error rate between the computed seconds per query (SPQ) and the actual value is as low as 1.7%.

Automatic Optimization of Hyperparameters for Deep Convolutional Neural Networks: Grid Search Enhanced with Coordinate Ascent

Sat, 03 Aug 2024 00:00:00 +0000

Convolutional neural network models of identical depth can yield significantly divergent results depending on the hyperparameter combination employed. Model performance can be improved by adjusting hyperparameter configurations; however, hyperparameter optimization typically incurs substantial computational resources and time. Thus, improving the efficiency of hyperparameter optimization is critical. This study adopts the coordinate ascent method, which provides only initial candidate values for each hyperparameter; at each iteration, only the single hyperparameter with the greatest influence on the model is altered, progressively expanding the search grid until accuracy converges. This method enables efficient and automated discovery of hyperparameter combinations that improve model accuracy. Experimental results show that, using the MWD dataset, the hyperparameter-optimized model achieves 95.71% accuracy on the validation set, and this hyperparameter combination can be regarded as an approximate global optimum. Furthermore, performance within the neighborhood of this combination proves stable, corroborating the robustness of the proposed hyperparameter optimization strategy.

Research on the Effectiveness of Different Outlier Detection Methods in Common Data Distribution Types

Sat, 27 Apr 2024 00:00:00 +0000

Outlier detection is widely applied in domains such as network performance optimization and machine learning data preprocessing. In machine learning, its objective is to improve data quality, thereby enhancing the performance of subsequent statistical analysis or machine learning models. Numerous effective and reliable outlier analysis methods exist, yet their effectiveness varies significantly across different data distribution types. Consequently, selecting an appropriate outlier analysis method is critical. This study conducts outlier detection on sample data from five continuous probability distributions (normal, chi-square, exponential, gamma, and t distributions) and four discrete probability distributions (binomial, Poisson, geometric, and hypergeometric distributions). Five outlier detection methods are employed — Z-Score, IQR, DBSCAN, Isolation Forest, and Random Forest — and their detection effectiveness is assessed. Through comparison and analysis, the characteristics exhibited by each outlier detection method when processing sample data from different distribution types are summarized. These findings will facilitate more informed method selection when confronted with diverse outlier detection scenarios.

Disease Prevention and Assistance System for Public Welfare Organizations Based on Lorentz-RR Analysis Technology

Thu, 07 Mar 2024 00:00:00 +0000

Health concerns have become a central focus in daily life. With the growing demand for human health monitoring, certain social organizations face an urgent need to strengthen health detection technologies, giving rise to various related techniques. This study aims to construct a disease prevention and assistance system for social organizations that integrates the strengths and resources of multiple parties — including social organizations, social service agencies, social workers, caregivers, service recipients, and their families. The main system adopts a front-end and back-end decoupled architectural design, underpinned by core technological innovation: namely, the Lorenz-RR scatter plot classification algorithm, encompassing classification algorithm selection, AlexNet algorithm development, and dataset expansion algorithm optimization, thereby completing the improvement of the AlexNet-based Lorenz-RR scatter plot classification algorithm and establishing the disease prevention and assistance system for social organizations. The system demonstrates high accuracy and high sensitivity in the analysis of service recipient indicators, offering substantial application value and broad potential for widespread adoption.

Participation in China International College Students Innovation Competition (2023) — Main Track

Wed, 06 Dec 2023 12:00:00 +0000

In 2023, as an alumnus of Suzhou University of Science and Technology, served as project lead for ‘Digital-School Think Tank — Digital Transformation Solution Provider for Industrial Enterprises’ and participated in the China International College Students Innovation Competition (Main Track), receiving a National Gold Award. The entire engagement centered on the capability gap issue in the digital transformation of small and medium-sized manufacturing enterprises.

Participation in the 5th China Young Entrepreneurs (Gongqingcheng) Development Summit

Wed, 24 Aug 2022 12:00:00 +0000

In 2022, participated in exchanges on enterprise digital technology and transformation processes, attending all three formal exchange components of the Summit: thematic seminars, industry dialogues, and technology feasibility assessments. Analyzed the high transformation failure rate among SMEs through causal attribution, proposed an operationalized reconstruction of the concept of ‘service,’ and advanced a functional repositioning of intellectual property.

Multi-Terminal Health Monitoring Platform Data Processing System for Complex Access Control Scenarios

Tue, 10 May 2022 12:00:00 +0000

Multi-Terminal Health Monitoring System Based on HRV and Lorenz-RR Analysis Techniques

Fri, 06 May 2022 12:00:00 +0000

A Hierarchical Software Design Methodology for High-Concurrency and Fine-Grained Access Control Scenarios

Fri, 22 Oct 2021 00:00:00 +0000

Complex software frequently faces high-concurrency and complex access control scenarios, and the development of access control modules in such contexts suffers from low code reuse rates and structural disorganization of functional modules. Providing a software design methodology for such scenarios is therefore vital for improving software productivity and quality. This paper proposes a hierarchical software design methodology for complex software development under high-concurrency and fine-grained access control scenarios. The methodology partitions the hierarchical structure according to the characteristics of the access control workflow, supporting fine-grained and multi-level permission control. For high-concurrency scenarios, a cache management layer is introduced to enhance the execution efficiency of the access control workflow. The specific layers within the methodology are delineated according to the practical requirements of software development. By decoupling user state detection, user state verification, and permission authentication within the access control workflow, the methodology ensures applicability across different framework environments and reduces security risks arising from high inter-framework coupling. Through a series of concurrency tests ranging from 0 to 500K, performance data across different frameworks under the same scenario are obtained, providing practitioners with a reference for selecting higher-performing frameworks based on specific deployment contexts.

An Omnidirectional Deep Learning Image Recognition Model Training Sample Acquisition System

Thu, 07 Oct 2021 12:00:00 +0000

Invited Talk at the 2nd International Conference on Computer Vision, Communications and Multimedia (ICCVCM 2021)

Sat, 21 Aug 2021 12:00:00 +0000

In 2021, delivered an invited talk entitled ‘Database Load Analysis Method for Query-Intensive Management Information Systems’ at the 2nd International Conference on Computer Vision, Communications and Multimedia (ICCVCM 2021). The talk proposed a database load measurement method for query-intensive management information systems, defining database load as the normalized ratio of the system’s current resource consumption state relative to its baseline performance boundary, expressed as a percentage.

A Timestamp-Based Reversible Encryption Algorithm for Access Control Passwords

Tue, 15 Jun 2021 12:00:00 +0000

A Timestamp-Based Reversible Encryption Algorithm for Access Control Passwords

Tue, 15 Jun 2021 12:00:00 +0000

ACO-Based Optimal Path Computation Software for Intelligent Warehousing

Tue, 15 Jun 2021 12:00:00 +0000

Customized Data Mining and Analysis System Based on KNN-CFA and CSA

Tue, 15 Jun 2021 12:00:00 +0000

A Clustered Computer Software Stability Testing System for Long-Term Operation

Thu, 10 Jun 2021 12:00:00 +0000

A Deep Learning-Based Image Object Recognition Model Training System

Thu, 03 Jun 2021 12:00:00 +0000

Distributed Enterprise Business Resource Management System Data Analysis and System Monitoring Software

Mon, 18 Jan 2021 12:00:00 +0000

Curriculum Vitae

Mon, 01 Jan 0001 00:00:00 +0000

Focuses on performance time series analysis, statistical modeling, and industrial fault prognostics and health management. Currently serves as Member of the Big Data Committee, Jiangsu Computer Society, and Member of the International Association of Applied Science and Technology. Reviewer for SCI/EI-indexed journals including Big Data and Internet Technology Letters. Has published multiple papers in the areas of performance time series modeling and interpretability, and holds 11 granted patents. Has led several municipal and enterprise-funded research projects, with outcomes successfully deployed in cloud computing monitoring and industrial control anomaly diagnosis scenarios.

Research Platforms

Mon, 01 Jan 0001 00:00:00 +0000

Belarusian State University

The Faculty of Applied Mathematics and Computer Science (FPMI), Belarusian State University, was established in 1970 as an institution for fundamental research and talent cultivation, and was among the first faculties of its kind founded during the former Soviet era. The Faculty encompasses two legally independent institutes — the Research Institute of Information Technology and Management and the Research Institute of Applied Mathematics Problems — forming an institute-faculty collaborative organizational structure.