【摘要】 当前物流业中，对于配送人员的薪酬计算大都基于配送距离和物品重量等因素，其缺乏对配送人员具体活动类型及能量消耗的考虑，难以对薪酬进行高效合理分配。基于此，在数据层面，通过与某大型物流公司合作，为25名配送人员穿戴相应设备，采集其在配送过程中的加速度计和陀螺仪等真实数据。算法层面，提出了一种基于多任务双向长短时记忆（BiLSTM）的深度网络结构，通过大量实验表明，BiLSTM模型在活动识别和能量消耗分级上的分类准确率分别达到92.8%和94.2%，结果皆优于基准多任务LSTM算法和其他代表性学习算法。更多还原

【Abstract】 In the current logistics industry, the salary calculation for delivery personnel is mostly based on factors such as delivery distance and weight of items, which lacks consideration of specific activity types and energy consumption of delivery personnel, making it difficult to allocate salary efficiently and reasonably. Based on this, in this paper, we cooperate with a large logistics company to collect real data such as accelerometer and gyroscope from 25 delivery personnel who wear wearable devices during the delivery process. At the algorithm level, a deep network structure based on multitask bidirectional long short-term memory(BiLSTM)is proposed. Extensive experiments show that the BiLSTM model achieves 92.8% and 94.2% classification accuracy for activity recognition and energy consumption classification, outperforming both the benchmark multitask LSTM algorithm and other representative learning algorithms.更多还原