Research on Temperature Drift Error Compensation for Electronic Balances Using Multi-RBF Neural Networks

To overcome the limitations of a single neural network model in accurately matching complex nonlinear processes, this paper introduces a temperature drift compensation method for electronic balances using a multi-RBF neural network model. Each sub-network of the multi-RBF neural network model introduced in this paper is dedicated to a specific calibration weighing segment of the electronic balance. The corresponding sub-RBF neural network is trained using data from its designated weighing segment, enabling the multi-RBF neural network model to characterize and match the complex global system properties. The paper primarily discusses the structure and learning algorithm of the multi-RBF neural networks, with testing and simulation analysis conducted in Matlab. Experimental results demonstrate that the proposed method surpasses traditional models across various real-world public datasets. Specifically, the multi-RBF neural network achieves root mean square errors of 0.0002 and 0.0004 in the first and second weighing stages, respectively, compared to errors of approximately 0.0008 and 0.0010 for the single neural network in these stages; These comparative results highlight the superior error fitting capabilities of the multi-network model over the single model, further enhancing measurement accuracy.At the end of the article, the shortcomings of the current stage were analyzed, providing research directions for the next stage of research.