Vehicle Weighing Compensation Algorithm Based on ECDOA-RGBPNN

This study proposes a novel vehicle weighing compensation algorithm integrating the Extended Collective Decision Optimization Algorithm (ECDOA) with a Residual Gated Backpropagation Neural Network (RGBPNN) to address measurement inaccuracies caused by multidimensional interference factors in Weight-in-Motion (WIM) systems. The methodology enhances conventional backpropagation neural networks (BPNN) by incorporating residual connections and gating mechanisms to construct RGBPNN, improving nonlinear modeling between input features and target values. During training, ECDOA is used to initialize the weight matrix and bias vector of RGBPNN to alleviate the problem of premature convergence caused by random initialization. Experimental validation demonstrates an RMSE of 1,285 kg, representing improvements of 8.52%, 5.43%, and 0.73% over He-BPNN, GA-BPNN, and MEA-BPNN baselines, respectively, with superior MAPE and R2 performance. Ablation analysis confirms three contributions: ECDOA initialization alone reduces RMSE by 7.55%, residual connections enhance training stability, and gating mechanisms provide an additional 0.97% error reduction. The ECDOA-RGBPNN framework enables systematic WIM error correction, offering a novel solution for non-stop vehicle weighing detection with applications in logistics, transportation, and highway management systems.