军事知识服务平台
基于FPGA模拟双类型忆阻器的智能驾驶转向控制研究
Research on FPGA-based dual-type memristor simulation for intelligent driving steering control
摘要 提出一种基于FPGA模拟双忆阻器的智能驾驶方向控制系统,设计忆阻器硬件电路,构建忆阻器脉冲神经网络硬件模拟系统以实现网络分类功能;权重忆阻器承担卷积层乘法运算与SNN全连接层权重乘法任务,通过电导动态调节实现学习和加权计算;泄漏积分发放(LIF)脉冲忆阻器集成于SNN输出层,模拟生物神经元"积分-点火"机制,生成精准尖峰信号.基于自建数据集进行训练,并在Xilinx Zynq 7020 FPGA硬件平台上进行测试,结果显示车辆转向任务正确率达92%,转向控制忆阻器脉冲网络系统功耗仅为1.768 W,有效提高了轻量级忆阻器硬件模拟网络的识别精度.该工作验证了忆阻器模型在脉冲神经网络识别应用中的有效性,为智能驾驶方向控制提供了一种兼具实时性、低功耗与高可靠性的忆阻器创新硬件神经网络验证参考方案.
Abstract:
This study proposes a dual-type memristor-based SNN steering control system implemented on FPGA.The memristor hardware circuits are designed to construct a memristor-based spiking neural network(SNN)hardware simulation system that performs network classification tasks.Weight memristors are employed to execute multiplication operations in convolutional layers and weight multiplication in fully connected SNN layers,enabling learning and weighted computation through dynamic conductance modulation.A leaky integrate-and-fire(LIF)spiking memristor is integrated into the output layer of the SNN,mimicking the biological neuronal"integrate-and-fire"mechanism to generate precise spike signals.The system is trained using a self-constructed dataset,and experimental results on the Xilinx Zynq 7020 FPGA hardware platform validate the effectiveness of the proposed memristor model in SNN-based recognition applications.This work offers a novel memristor-based hardware neural network reference solution for intelligent driving steering control,characterized by real-time performance,low power consumption,and high reliability.
年,卷(期): 2026, 58(3)
分类号: TP183
机标分类号: TP391.41 TP183 TP273
在线出版日期: 2026年10月15日
基金项目: 2023YFB440230:国家科技计划项目52472149:国家自然科学基金20260201080GX:吉林省科技发展计划重点研发项目,