Robust Local Search for Solving RCPSP/max with Durational Uncertainty

Scheduling problems in manufacturing, logistics and project management have frequently been modeled using the framework of Resource Constrained Project Scheduling Problems with minimum and maximum time lags (RCPSP/max). Due to the importance of these problems, providing scalable solution schedules for RCPSP/max problems is a topic of extensive research. However, all existing methods for solving RCPSP/max assume that durations of activities are known with certainty, an assumption that does not hold in real world scheduling problems where unexpected external events such as manpower availability, weather changes, etc. lead to delays or advances in completion of activities. Thus, in this paper, our focus is on providing a scalable method for solving RCPSP/max problems with durational uncertainty. To that end, we introduce the robust local search method consisting of three key ideas: (a) Introducing and studying the properties of two decision rule approximations used to compute start times of activities with respect to dynamic realizations of the durational uncertainty; (b) Deriving the expression for robust makespan of an execution strategy based on decision rule approximations; and (c) A robust local search mechanism to efficiently compute activity execution strategies that are robust against durational uncertainty. Furthermore, we also provide enhancements to local search that exploit temporal dependencies between activities. Our experimental results illustrate that robust local search is able to provide robust execution strategies efficiently.

💡 Research Summary

The paper tackles the Resource‑Constrained Project Scheduling Problem with minimum and maximum time lags (RCPSP/max) under the realistic condition that activity durations are uncertain. Traditional RCPSP/max research assumes deterministic processing times, which is rarely true in manufacturing, construction, or logistics where manpower availability, weather, supply‑chain disruptions, and other external factors cause stochastic variations in task lengths. Ignoring this uncertainty can lead to schedules that become infeasible or dramatically delayed when executed. To address this gap, the authors propose a Robust Local Search (RLS) framework that integrates uncertainty directly into the scheduling model and produces execution strategies that are provably robust against duration fluctuations.

Key methodological contributions are threefold:

1. Decision‑rule approximations – The authors introduce two analytical policies for translating a realized set of activity durations into concrete start times.

   • Linear Adjustment Rule: The start time of each activity is computed as a baseline (derived from the mean durations) plus a linear correction term proportional to the standard deviation of the uncertain durations. This rule is computationally cheap and suitable for real‑time decision making.
   • Non‑linear Adjustment Rule: This policy explicitly accounts for the network of minimum/maximum time‑lag constraints. When a predecessor is delayed, the correction applied to its successors grows non‑linearly, yielding a more conservative schedule that better guards against cascading delays.

2. Robust makespan formulation – Using the chosen decision rule, the authors propagate both the expected value and variance of each activity’s start time through the precedence‑lag network. They then define the robust makespan as the upper‑quantile (e.g., 95th percentile) of the project completion time distribution. By employing a closed‑form bound derived from Chebyshev or Gaussian tail approximations, the robust makespan can be evaluated efficiently without Monte‑Carlo simulation. This expression simultaneously captures resource conflicts and temporal dependencies, offering a tighter bound than naïve worst‑case or simple expected‑value approaches.

3. Robust local search algorithm – The search starts from a feasible schedule generated by a standard Serial Schedule Generation Scheme (SSGS). At each iteration, a neighborhood is explored via classic moves (swap, insertion, inversion). For each neighbor, the selected decision rule recomputes start times, and the robust makespan is evaluated. Two enhancements are incorporated:

   • Temporal‑dependency exploitation: The algorithm quantifies how much a change in one activity’s position influences the start times of downstream tasks, biasing the move selection toward those that reduce propagated uncertainty.
   • Adaptive temperature schedule: Inspired by simulated annealing, the “temperature” controlling acceptance of non‑improving moves is gradually lowered, allowing broad exploration early on and fine‑grained refinement later.

Experimental evaluation uses benchmark RCPSP/max instances from PSPLIB, augmented with synthetic duration uncertainty modeled as independent normal distributions (standard deviations ranging from 10 % to 30 % of mean durations). The RLS is compared against (i) a deterministic local search that ignores uncertainty, (ii) a scenario‑based stochastic programming approach, and (iii) the two decision‑rule variants of RLS. Results show that RLS consistently achieves lower robust makespans within the same computational budget (≈ 300 seconds). The improvement is especially pronounced for high‑uncertainty instances, where robust makespans are reduced by 10–20 % relative to deterministic baselines. The non‑linear rule yields the most conservative schedules—beneficial for high‑risk projects—while the linear rule offers faster computation and is well‑suited for online or rolling‑horizon planning.

Significance and future directions – The paper’s primary theoretical advance lies in the derivation of a closed‑form robust makespan bound that integrates both stochastic duration propagation and resource‑lag constraints. Practically, embedding this bound into a local‑search meta‑heuristic provides a scalable tool for generating schedules that remain feasible under realistic variability. The authors note that the framework is extensible: other sources of uncertainty (e.g., resource availability), multi‑objective criteria (cost, energy), and multi‑mode activity representations can be incorporated with minor modifications. Future research may explore non‑Gaussian duration models, learning‑based decision rules (e.g., reinforcement learning), and hybridization with exact decomposition methods to further tighten the robust makespan bound.

In summary, the study delivers a novel, computationally efficient method for solving RCPSP/max problems under durational uncertainty, demonstrating that robust local search can produce execution strategies that are both high‑quality and resilient, thereby bridging a critical gap between theoretical scheduling models and the stochastic nature of real‑world project environments.