Zero-Drift Error Compensation Method for Electronic Balance Based on Incremental Broad Learning Network

The zero-drift error is an important indexes for evaluating the performance of electronic balance, which directly affects the weighing accuracy of electronic balance. This article proposes a compensation method for zero-drift error of electronic balance based on a broad learning network (BLN) and constructs the compensation framework of zero-drift error. Due to different zero-drift errors of the electronic balance when no-loaded and loaded, by using the historical zero error data, a combined error compensation model based on incremental BLN is established to compensate the zero-drift errors in the above two cases, and the detailed training algorithm is provided. Unlike deep neural networks, BLN only needs to partially adjust the feature nodes and enhancement nodes to correct the error compensation model online and reduce computational complexity. This proposed method based on incremental BLN is used to test the compensation of electronic balance with a range of 200 g/1 mg in the field. The experimental results show that the standard deviation of the weighing results within 15 min is stable within 1 mg, and that of the weighing results within 5 h is stable within 5 mg. The zero-drift error is effectively compensated, and it is far smaller than the weighing error before compensation, which confirms the effectiveness of this proposed method.