Multi-Objective Optimization

A solution is Pareto optimal if no objective can be improved without degrading at least one other objective. In grid reconfiguration, this means you cannot reduce line losses further without increasing the number of switching operations or violating a radiality constraint. The set of all non-dominated solutions forms the Pareto front, giving network planners a trade-off curve rather than a single answer. This is critical because a single 'best' topology rarely exists when balancing SAIDI reliability metrics against operational costs.

The weighted sum method scalarizes multiple objectives into a single aggregate function by assigning a weight coefficient to each goal. For a Distribution Feeder Reconfiguration (DFR) problem, the objective might be w1 * (Power Loss) + w2 * (Switch Operations). While computationally simple and compatible with Mixed-Integer Linear Programming (MILP) solvers, this approach has a critical limitation: it cannot discover solutions on non-convex regions of the Pareto front. Varying weights systematically is required to approximate the full trade-off space.

This technique optimizes a single primary objective while treating all other objectives as constraints bounded by epsilon values. For Service Restoration (SR), you might minimize Cold Load Pickup (CLPU) duration while constraining the number of Intelligent Electronic Device (IED) operations to ≤ 5. By iteratively tightening the epsilon bounds, the method generates the full Pareto front, including non-convex regions that the weighted sum method misses. This is particularly valuable for N-1 Criterion contingency planning where hard operational limits exist.

Algorithms like NSGA-II (Non-dominated Sorting Genetic Algorithm II) and MOEA/D use population-based heuristics to evolve a diverse set of Pareto-optimal solutions in a single run. These are well-suited for Network Reconfiguration because they handle the discrete, non-linear nature of switch statuses without gradient information. Key mechanisms include:

• Non-dominated sorting: Ranking solutions by dominance fronts
• Crowding distance: Preserving diversity along the Pareto front
• Crossover and mutation: Operators adapted for radiality constraint preservation They scale effectively to many-objective problems with three or more competing goals.

Goal programming defines aspiration levels for each objective and minimizes the weighted sum of deviations from these targets. In a Volt-VAR Optimization context, the target might be 0.95 per-unit voltage with zero Conservation Voltage Reduction (CVR) violations. The solver minimizes both over-achievement and under-achievement deviations. This approach is intuitive for utility engineers because it mirrors operational planning: 'I want losses below 3% and switching operations under 10.' It integrates naturally with Model Predictive Control (MPC) frameworks for dynamic reconfiguration.

This technique identifies a single 'best-compromise' solution by measuring distance from an ideal but infeasible utopia point where all objectives are simultaneously at their individual optima. Using Lp-metrics (e.g., Euclidean distance for p=2), it finds the Pareto-optimal solution closest to this ideal. For Feeder Load Balancing, the utopia point might represent zero losses and perfect balance. Compromise programming is valuable when a decision-maker needs one actionable topology rather than an entire Pareto front for Distribution Automation (DA) execution.

The set of non-dominated solutions where improving one objective necessarily degrades another. In grid reconfiguration, a solution is Pareto optimal if no other topology can simultaneously achieve lower losses and fewer switching operations.

• Dominance: Solution A dominates B if A is better in at least one objective and no worse in all others
• Visualization: Plotted as a trade-off curve in 2D or a surface in 3D objective space
• Decision-making: The front presents operators with a menu of mathematically optimal compromises
• Knee point: The solution on the front where a small sacrifice in one objective yields a large gain in another

A scalarization technique that combines multiple objectives into a single aggregate function by assigning relative importance weights. The optimization then minimizes w₁ × Losses + w₂ × Switching Operations subject to radiality and voltage constraints.

• Weight selection: Requires domain expertise to set meaningful trade-off ratios
• Limitation: Cannot discover solutions on non-convex regions of the Pareto front
• Sensitivity analysis: Running with varied weight vectors generates an approximation of the optimal front
• Normalization: Objectives must be scaled to comparable magnitudes to avoid bias

An alternative scalarization approach that optimizes a single primary objective while treating others as constraints bounded by epsilon values. For example, minimize power losses subject to the constraint that switching operations ≤ ε.

• Advantage: Can identify solutions on non-convex Pareto fronts that weighted sum methods miss
• Grid application: Set a maximum allowable switching budget and find the lowest-loss topology within that budget
• Lexicographic variant: Prioritize objectives in strict hierarchical order
• Implementation: Solve a sequence of single-objective MILP problems with progressively tightened epsilon bounds

Population-based metaheuristics such as NSGA-II and MOEA/D that evolve a diverse set of Pareto-optimal solutions simultaneously. Well-suited to the combinatorial, non-linear nature of distribution network reconfiguration.

• NSGA-II: Uses non-dominated sorting and crowding distance to maintain diversity
• Encoding: Switch statuses represented as binary chromosomes in the genetic algorithm
• Constraint handling: Radiality enforced through repair operators or penalty functions
• Advantage over MILP: Handles non-linear objective functions like minimizing customer interruption minutes alongside technical losses

Selects a single preferred solution from the Pareto front by minimizing the distance to an ideal point—a hypothetical solution that simultaneously achieves the best possible value for every objective.

• Distance metrics: L₁ (Manhattan), L₂ (Euclidean), or L∞ (Chebyshev) norms define different compromise philosophies
• Chebyshev norm: Minimizes the worst deviation from the ideal, ensuring balanced underachievement across all objectives
• Grid context: Find the topology that comes closest to simultaneously achieving zero losses and zero switching operations
• Weighted variant: Incorporate operator preferences into the distance calculation

The post-optimization process of selecting a single implementation solution from the Pareto front using structured decision-maker preferences. Techniques include TOPSIS, AHP, and fuzzy logic ranking.

• TOPSIS: Ranks solutions by simultaneous proximity to the ideal and distance from the anti-ideal
• AHP: Decomposes the decision into a hierarchy and derives weights through pairwise comparisons
• Fuzzy MCDM: Handles linguistic preferences like 'slightly more important' when precise weights are unavailable
• Operator integration: Allows control room engineers to apply operational context that mathematical models cannot capture