Logo
0

升舱逻辑流程.excalidraw

优化模型:

参数定义

参数/变量定义
Pi(Ac)P_{i}(A\mid c)订单ii预分单接受度(不补贴)
PiupP_{i}^{up}订单ii升舱完单率(默勾ARAR模型计算)
PibaseP_{i}^{base}订单ii原本完单率(默勾ARAR模型计算)
priceiupprice_{i}^{up}订单ii升舱后价格(by轮询商家)
priceibaseprice_{i}^{base}订单ii勾选的券后最高价
ziz_{i}订单ii是否升舱(by轮询商家)
θ\theta补贴率
GMV0GMV_{0}历史大盘GMVGMV

模型

maxf(Z)=i=1Nzi(Piup[Pi(Ac)Piup+(1Pi(Ac))Pibase])\begin{gather} maxf(\mathbf{Z})=\sum_{i=1}^{N}{{z_{i} (P_{i}^{up}-[P_{i}(A|c) P_{i}^{up}+(1-P_{i}(A|c)) P_{i}^{base}])}} \end{gather} s.ti=1Nzi(priceiuppriceibase)PiupBB=GMV0θ\begin{gather} s.t \quad \quad \sum_{i=1}^{N}{z_{i}(price_{i}^{up}-price_{i}^{base})P_{i}^{up}} \leq B \\ B=GMV_{0} \cdot \theta \end{gather}

所需模型

  1. 预估完单率模型:使用默勾ARAR的结果计算:[https://ether.intra.hongyibo.com.cn/deploy/466] ARAR模型: 日志表:kflower_strategy.ods_log_kflower_ether_price_anchor_cr_model 特征:城市、bubble_id、pid、起终点经纬度(flng,flat,tlng,tlat)、运力品牌(product_id)、预估价格(pre_total_fee)、order_id、应答率类型(answer_rate_type)?、third_party_ft?,real_time_ft?,trace_id,estimate_id,订单表单商家数量(product_cnt)

  2. 预分单接受度模型 由于升舱判断前置,故拿不到接驾距离,需要重新训练模型 ❗注意等待时长、预分单价格的单调关系

求解

Pasted image 20240901111759.png

τir=(1Pi(Ac))(PiupPibase)\tau_{i}^{r}=(1-P_{i}(A \mid c))(P_{i}^{up}-P_{i}^{base}) τic=(priceiuppriceibase)Piup\tau_{i}^{c}=(price_{i}^{up}-price_{i}^{base})P_{i}^{up}

© 2025 All rights reservedBuilt with DataHub Cloud

Built with LogoDataHub Cloud