大型衡器偏载误差机理与多传感器称量融合方法研究

Large-scale weighing instrument(LSWI) is the key measurement apparatus on the field of intelligent traffic, freight forwarding and logistics, etc, and it is a typical nonlinear and badly coupled multi-sensor system. The eccentric error compensation of the existed LSWIs is based on manually adjusting method, which has the disadvantages of difficult compensation on the eccentric error, hard labor, and low accuracy of weighing result. This is the urgent problem of static weighing filed. The aim of this project is to study eccentric error mechanism and weighing fusion method for LSWI, and supply the way of solving this problem in LSWI. The major works include: ①the study of LSWI's eccentric error mechanism by finite-element method, and the exploration of the eccentric error law and multi-sensor correlation, which is the theory basis of eccentric error compensation; ②the construction of constraint condition based the prior knowledge of LSWI and the optimum design method for improved raised-cosine radial basis fuction(RC-RBF) neural network ensemble, and then the establishment of the weighing fusion model of LSWI, which can perform the eccentric error automatically compensating and the LSWI accurately weighing; ③the construction of weighing experimental platform based on multiple load cells and the simulation and testing of weighing fusion and eccentric error compensation method, and then the optimization of the model's parameters. The project will present the theory and method of LSWI's eccentirc error compensation and weighing fusion, and solve the problem of difficult compensation and low accuracy in the existed LSWI. This proposed method is useful to detect and estimate the signal and compensate the errors.of mesurement and control system with multiple sensors, and this project will provide a new idea of nerual network ensembles optimun design.

大型衡器是智能交通、货运物流等质量称量的关键计量器具，是典型的非线性、强耦合性多传感器系统。现有大型衡器通过人工调节偏载误差，工作量大、补偿效果差、称量准确度低，是国内外静态称重领域亟待解决的难题。本项目研究大型衡器偏载误差机理与多传感器称量融合方法，主要包括：①基于有限元法的偏载误差机理研究，揭示偏载误差规律与多传感器关联性，为误差补偿提供理论依据；②构造基于先验知识的约束条件，建立改进型升余弦-径向基函数(RC-RBF)神经网络集成优化方法，在此基础上构建大型衡器偏载误差补偿与称重信息融合模型，实现偏载误差自动补偿与准确称重；③构建多传感器称重模拟实验平台，验证理论方法的准确性与实用性。项目研究可形成大型衡器偏载误差补偿与称量融合理论与方法，解决现有大型衡器存在的严重问题，关键技术与方法对其它多传感器测控系统的误差补偿与检测估计具有重要借鉴意义，也为神经网络集成优化研究开辟一种新思路。

大型衡器（汽车衡、轨道衡等）是智能交通、货运物流等质量称量的关键设备，是典型的非线性、强耦合性多传感器系统。偏载误差是造成大型衡器称量误差的主要原因，目前采用的人工调节法、线性回归法、角差修正法等，都将大型衡器简单等效为多输入单输出的线性系统，误差补偿效果差、效率低，严重影响了称量结果的准确性与可靠性，是目前大型衡器称重计量领域亟待解决的难题之一。针对这一问题，本项目以汽车衡为研究对象，主要开展了如下研究工作：1）通过分析汽车衡偏载误差影响因素和误差来源，深入研究了汽车衡偏载误差机理，并开展了基于有限元法的汽车衡受力分析方法研究。2）研究了基于先验知识的神经网络优化方法，形成了一套适合小样本情况下的神经网络优化设计方法。通过提炼系统先验知识（包括系统的单调性、有界性、光滑性、系统误差分布特征）、研究先验知识对神经网络优化的影响机制，构建了神经网络优化的约束条件，建立了条件约束下的神经网络训练算法，分析或证明了算法的收敛性，有效提高了小样本情况下神经网络模型的泛化能力；在此基础上，建立了基于理想称重模型与神经网络集成的大型衡器称重融合方法、基于导数约束的称重传感器非线性误差补偿方法、基于惩罚函数与系统误差分布特性的汽车衡称重融合方法、基于神经网络权值光滑约束的汽车衡称重融合方法等。3）构建了DSP+8路ADC的多传感器称重实验平台，验证了理论方法与模型的可行性，同时为大型衡器下一步研究提供了试验平台。本项目研究形成的大型衡器偏载误差补偿与称重融合新方法，大大提高了称重结果准确性；同时形成的神经网络优化方法为小样本情况下神经网络优化设计提供了一种新思路，对其它测控系统建模、误差补偿具有借鉴作用。