张伟, 付良超, 吴中华,等.基于ATCSO和神经元竞争的SOFNN设计[J].河南理工大学学报（自然科学版）,2026,45(3):10-20.

ZHANG W, FU L C, WU Z H,et al.Self-organizing fuzzy neural network based on attractor triple competitive swarm optimization （ATCSO） and neuron competition[J].Journal of Henan Polytechnic University(Natural Science) ,2026,45(3):10-20.

基于ATCSO和神经元竞争的SOFNN设计

张伟1,2, 付良超1, 吴中华1,2

1.河南理工大学 电气工程与自动化学院，河南 焦作  454003；2.河南理工大学 河南省煤矿装备智能检测与控制重点实验室，河南 焦作 454003

摘要:目的 针对模糊神经网络结构调整缺乏可解释性和网络参数优化精度不足等问题，提出一种基于神经元竞争机制和吸引子的三重竞争群优化算法的自组织模糊神经网络（self-organizing fuzzy neural network based on attractor triple competitive swarm optimization and neural competition， NCSOFNN-ATCSO）设计方法。  方法 首先，提出一种基于神经元竞争的网络调整机制，赋予结构调整以生物学意义；其次，设计神经元竞争力指标与Axin2基因表达水平，并采用单边Jacobi分解规则层神经元输出矩阵，以准确量化各神经元对于网络的重要性，提高神经元竞争的准确性；最后，为提高网络预测精度，采用基于动态吸引子的三重竞争群优化算法优化网络参数，引入三重竞争机制提升网络优化速度，并设计动态吸引子，以找到更优的参数向量。  结果 通过基准测试函数验证ATCSO算法性能，所提算法效率和精度更高；通过时间序列预测实验测试所提网络模型，相较于各对比模型，NCSOFNN-ATCSO精度更高且结构更精简。此外，将所提网络模型应用于出水氨氮的质量浓度预测，能够较为准确地预测出水氨氮质量浓度。  结论 与其他网络模型相比，所提NCSOFNN-ATCSO能够得到结构紧凑且预测精度较高的网络模型。

关键词:模糊神经网络；神经元竞争；单边Jacobi；三重竞争机制；动态吸引子

doi:10.16186/j.cnki.1673-9787.2024100001

基金项目:国家自然科学基金资助项目（61903126）；河南省科技攻关项目（222102210213）

收稿日期:2025/02/12

修回日期:2025/04/21

出版日期:2026/04/28

Self-organizing fuzzy neural network based on attractor triple competitive swarm optimization （ATCSO） and neuron competition

Zhang Wei1,2, Fu Liangchao1, Wu Zhonghua1,2

1.School of Electrical Engineering and Automation， Henan Polytechnic University， Jiaozuo  454003， Henan， China；2.Henan Key Laboratory of Intelligent Detection and Control of Coal Mine Equipment， Henan Polytechnic University， Jiaozuo  454003， Henan， China

Abstract: Objectives To address the issues of poor interpretability in the structural adjustment of fuzzy neural networks （FNNs） and insufficient accuracy in parameter optimization， a self-organizing fuzzy neural network based on neuron competition and attractor triple competitive swarm optimization （NCSOFNN-ATCSO） is proposed.  Methods First， a neuron competition-based structural adjustment mechanism is developed to endow the network evolution process with biological interpretability. Second， a neuron competitiveness index incorporating Axin2 gene expression level is constructed. Meanwhile， the output matrix of rule-layer neurons is decomposed using the one-sided Jacobi method to accurately quantify neuron importance and improve the precision of neuron competition. Finally， to enhance prediction accuracy， network parameters are optimized using a triple competitive swarm optimization algorithm with dynamic attractors （ATCSO）. A triple competition mechanism is introduced to accelerate convergence， and dynamic attractors are designed to obtain superior parameter vectors.  Results The performance of ATCSO is validated using benchmark functions， demonstrating superior efficiency and accuracy. In time series prediction tasks， the proposed NCSOFNN-ATCSO achieves higher accuracy and a more compact structure compared with other models.  When applied to effluent ammonia nitrogen concentration prediction， the model provides accurate estimation results.  Conclusions The proposed NCSOFNN-ATCSO achieves a compact structure and higher prediction accuracy than existing network models.

Key words:fuzzy neural network；neuron competition；one-sided Jacobi method；triple competition mechanism；dynamic attractor