Voltammetry is a promising technique for estimating heavy metal pollution such as Cadmium (Cd) and Lead (Pb) in water. Its advantages include rapid analysis and cost-effectiveness over established methods like Atomic Absorption Spectroscopy (AAS) and Inductively Coupled Plasma - Mass Spectrometry (ICP-MS). However, current analysis often depends only on peak data, ignoring the rest of the voltammetric signal which may contain useful information that could potentially improve measurement accuracy. To address this limitation, the Cross-Attention Feature Fusion (CAFF) network is proposed to analyze Cyclic Voltammetry (CV) signals acquired using a 3-electrode setup with a Glassy Carbon Electrode (GCE) as the working electrode, Platinum as the counter, and Ag/AgCl as the reference. Unlike standard self-attention mechanisms or simple concatenation fusion methods, CAFF introduces a novel dual-stream architecture that dynamically captures the inter-dependencies between raw CV signals and extracted peak data—an approach previously unexplored in electrochemical sensing. The model integrates an Improved Beluga Whale Optimization (IBWO) algorithm that automatically determines the optimal hyperparameters, resulting in a more robust model. Robustness was assessed using Chemically-Informed Degradation Simulation (CIDS). As a result, the proposed CAFF-IBWO model demonstrated superior performance, achieving R values of 0.97 for Cd and 1.00 for Pb. It also significantly reduced the Mean Absolute Percentage Error (MAPE) by 65.79% for Cd and 72.50% for Pb compared to single-input attention networks. Furthermore, CAFF-IBWO exhibited remarkable resilience against signal degradation, maintaining stable prediction performance across varying noise conditions. While the study focuses specifically on Cd and Pb and requires further validation for broader generalization, the demonstrated performance is highly promising. These findings underscore the model’s potential for real-world environmental sensing applications.
扫码关注我们
求助内容:
应助结果提醒方式:
