Abstract:We introduce a strategic decision-making problem faced by logistics providers (LPs) seeking facility location decisions that lead to profitable operations. The profitability depends on the revenue generated through agreements with shippers, and the costs arising when satisfying those agreements. The latter depend in turn on service levels and on characteristics of the shippers' customers. However, at a strategic level, LP has imperfect information thereof.
We propose a stochastic bilevel formulation where a given LP (leader) anticipates the decisions of shippers (followers) arising from a random utility maximization model. Using a sample average approximation and properties of the associated optimal solutions, we introduce a non-conventional single-level mixed integer linear programming formulation that can be solved by a general-purpose solver. We can quickly identify situations that lead to zero expected profit for the LP. Experimental results show that optimal expected profit is highly dependent on shippers' price sensitivity. Underestimating it can lead to an overestimation of expected profits.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is how logistics providers (LPs) can maximize their profits in facility location decisions. Specifically, logistics providers need to decide where to open facilities and what service levels and prices to offer in order to attract shippers to sign agreements while ensuring that the costs are not too high. The challenges of this problem are as follows:
1. **Profit maximization**: Profit depends on the revenue obtained from signing agreements with shippers and the costs incurred in fulfilling these agreements. Costs are also affected by service levels and the characteristics of shippers' customers.
2. **Incomplete information**: At the strategic planning level, logistics providers have incomplete information about the characteristics of shippers and their customers.
3. **Random utility maximization model**: Shippers' decisions are based on a random utility maximization model, which means that their choices are uncertain.
To address these challenges, the paper proposes a stochastic bi - level optimization model, in which the logistics provider is the leader and the shipper is the follower. Through sample average approximation (SAA) and the properties of the optimal solution, the paper further proposes an unconventional single - level mixed - integer linear programming (MILP) model that can be solved using general solvers.
### Main contributions
1. **Practical relevance**: A problem highly relevant to logistics providers has been studied and an attempt has been made to solve this problem for the first time.
2. **Stochastic bi - level model**: A stochastic bi - level optimization model has been proposed, in which shippers' decisions are based on a random utility maximization model, and the model includes endogenous and exogenous uncertainties in logistics providers' costs.
3. **Unconventional single - level MILP model**: Through simulation model approximation and the properties of the optimal solution, an unconventional single - level MILP model has been proposed.
4. **Experimental analysis**: The influence of model parameters on solutions and computation time has been analyzed through experiments. The results show that the optimal expected profit is highly dependent on shippers' price sensitivity. Underestimating price sensitivity may lead to an overestimation of the expected profit.
### Mathematical model
The mathematical model in the paper includes the following main parts:
1. **Objective function**:
\[
\max E_{\epsilon}[G(y, z, r, x, w) : (y, z, r, w) \in H(x); x \in F(y, z; \epsilon)]
\]
where \( G( \cdot ) \) is the objective function of the logistics provider, \( H( \cdot ) \) is the constraint condition, and \( F(y, z; \epsilon) \) is the optimal solution set of shippers.
2. **Optimal solution set of shippers**:
\[
F(y, z; \epsilon) = \arg\max \left[ \sum_{n \in N} \sum_{k \in K_n} \sum_{m \in M_0^{nk}} U_m^{nk}(y, \epsilon_m^{nk}) x_m^{nk} : x_m^{nk} \in R(z); n \in N, k \in K_n, m \in M^{nk} \right]
\]
3. **Feasible solution set of shippers**:
\[
R(z) = \left\{ \sum_{m \in M_0^{nk}} x_m^{nk} = 1, \forall n \in N, k \in K_n; x_m^{nk} \le z_m^{nk}, n \in N, k \in K_n, m \in M^{nk} \right\}
\]
4. **First - stage problem**:
\[
\max - \sum_{i \in I} f_i r_i + Q(r, z, y)
\]
Constraints include:
\[
\sum_{p \in P_m^n} y_p^{mn} \le 1, \quad n \in N, m \in M_n
\]