Pith. sign in

REVIEW 5 major objections 5 minor 42 references

Learning to Sample in Variable Neighborhood Search Algorithm for Urban Cable Routing Optimization

T0 review · 5 major / 5 minor · reviewed 2026-08-03 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Co-optimizing substation connectivity and road-constrained cable paths, with shared-trench costs internalized, cuts urban cable construction cost by roughly 30–50% compared with representative 2024 approaches.

desk verdict Solid engineering paper with an overstated headline: the 30–50% cost saving is a modeling effect, not an algorithmic breakthrough; fair comparisons show 1–8% gains. read the letter →

arxiv 2512.19321 v2 pith:DY6VVHYA submitted 2025-12-22 cs.NE

Wei Liu , Rui Wang , Chenhui Lin , Kaiwen Li , Wenhua Li , Tao Zhang This is my paper · ORCID
classification cs.NE
keywords urbancableroutingconnectivity–pathco-optimizationvariableneighborhoodsearchdeepreinforcementlearningmulti-agenttrenchsharingdistributionnetworkplanningbenchmarksuite
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Urban underground cable networks are extremely expensive, and traditional planning algorithms typically decide which substations to connect using straight-line distances, only afterwards drawing cable paths along roads. This paper argues that this sequential, relation-only approach costs cities 30–50% more than necessary, because in cities cables must follow roads and running several cables in the same trench sharply reduces construction cost. It reformulates the problem as a two-level co-optimization of substation connectivity and road-constrained routing, and solves it with a multi-operator variable neighborhood search whose neighborhood choices are guided by a multi-agent deep-reinforcement-learning module. On four synthetic grid benchmarks, the proposed L-MVNS is reported to cut total construction cost by roughly 30–50% versus representative 2024 baselines, with lower run-to-run variance on larger instances. If the result transfers to real street layouts, it gives planners a practical way to save a large fraction of an expensive infrastructure budget.

What carries the argument

The carrying mechanism is a two-level formulation: a substation-level graph encodes which feeders connect which substations, and a road-level graph encodes which road segments are trenched and how many parallel cables each carries. The algorithm iteratively destroys and repairs both levels: three destruction operators (path-related removal, intra-feeder 2-opt, and inter-feeder 2-opt) alter the candidate network, a modified A* repair re-plans affected paths while counting trench-sharing costs, and an adaptive perturbation size balances exploration and exploitation. In L-MVNS, three lightweight agents (each an LSTM plus multi-head attention) read the current solution and output a probability f

What would settle it

Run L-MVNS and a relation-only baseline (e.g., MCWS) on a real GIS road network with actual trenching and cable cost data for a mid-size city; if the cost gap is well below 30%, the headline savings do not generalize. Alternatively, on the paper's own Case-1 instance, set the trench-to-cable cost ratio to 1:1 and check whether the reported gap between path-aware and relation-only methods collapses—which would show the savings are tied to the cost assumption.

Watch

Extended reading notes

Core claim

The paper's central claim is that co-optimizing which substations to connect and exactly which road segments to excavate, with shared-trench savings internalized, reduces total construction cost by approximately 30–50% against representative 2024 planning approaches. The evidence comes from four synthetic regular-lattice benchmarks, where path-aware methods consistently beat relation-only baselines (MCWS and HGS) that use precomputed shortest paths and ignore trench sharing; the multi-operator variable neighborhood search (MVNS) and its learning-assisted version (L-MVNS) give the lowest costs, and L-MVNS adds moderate gains and markedly lower variance on larger cases.

Load-bearing premise

The claim that planning with explicit roads and shared trenches saves 30–50% rests on four synthetic regular-grid benchmarks with a fixed 3:1 trench-to-cable cost ratio standing in for real urban road networks; if those benchmarks are not representative, the savings figure may not transfer to actual cities.

Editorial extensions

If this is right

  • Using path-aware, trench-sharing planning instead of relation-only planning is reported to cut total construction cost by roughly 30–50% on the benchmark suite.
  • The multi-agent DRL guidance reduces run-to-run variance, making cost estimates from the algorithm more predictable on large instances.
  • The four regular-lattice instances form a standardized, scalable benchmark for future urban cable routing algorithms.
  • The auxiliary-task initialization (hybrid genetic search for connections, A* for paths) produces feasible starting points that accelerate the subsequent neighborhood search without special tuning.
  • The method's gap to the best single-operator search is small but consistent, indicating that combining complementary destruction operators is more effective than any one operator alone.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The 30–50% figure is measured against MCWS and HGS on synthetic grids; whether it transfers to real street networks with irregular geometry and non-uniform costs is untested in the body of the paper, even though the abstract promises GIS-derived instances that do not appear in the reported experiments.
  • The benefit likely depends on the trench-to-cable cost ratio: if the ratio approaches 1:1, the incentive to share trenches weakens and the gap between path-aware and relation-only planning should shrink—the paper's sensitivity analysis varies only algorithmic parameters, not the cost ratio.
  • The same connectivity–path co-optimization pattern, with trench sharing priced in, could be applied to other shared-infrastructure networks such as district heating, fiber optics, or water mains, where routing constraints and joint excavation costs dominate.
  • Because the DRL module's gains over the non-learning MVNS are modest (sub-2% on reported cases), a sharper test would be to evaluate whether a learned destruction policy generalizes across instance scales and cost structures without per-instance fine-tuning.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 5 minor

Summary. The paper formulates urban cable routing as a bilevel connectivity–path co-optimization problem on a road graph, with costs for trenching and cable laying where parallel cables can share a trench. It proposes MVNS, a multi-operator variable neighborhood search with three destruction operators and A* repair, and L-MVNS, a DRL-guided variant in which three agents are trained to bias neighborhood sampling. Experiments are run on four synthetic regular-lattice benchmark cases (Section 3.5) with 10 independent runs per method. The paper claims that MVNS and L-MVNS reduce total construction costs by approximately 30–50% relative to representative 2024 approaches, and that L-MVNS further improves accuracy and variance on larger instances.

Significance. The practical problem is relevant: accounting for road-constrained routes and trench sharing can substantially change planning outcomes. If the algorithmic claims were cleanly established, the DRL-guided neighborhood selection in L-MVNS would be a useful contribution. The paper also provides a reproducible benchmark suite, though only four synthetic cases are actually used. However, the headline 30–50% cost reduction is an artifact of comparing against baselines that optimize a different cost model; the fair path-aware comparison shows only 0.12–8% gains. The abstract also promises 12 benchmark and 3 GIS instances that do not appear in the body. These issues undermine the central claims as currently stated.

major comments (5)
  1. [Abstract, §5.2, Table 2] The claim that MVNS/L-MVNS reduce total construction costs by approximately 30–50% is not supported by the paper's own fair algorithmic comparisons. Table 2 shows that this large gap appears only against MCWS and HGS, which optimize the Stage-I objective (Eq. (2)) and explicitly ignore trench sharing. Against path-aware baselines SNS-1/2/3, which use the same cost model (Eq. (6)), MVNS/L-MVNS are only 0.12–8.00% better (Case-2: 0.12%, Case-4: 0.18%). The 30–50% figure is therefore a modeling effect of allowing shared trenches, not an algorithmic improvement. The abstract and conclusion should be reframed to separate the modeling benefit from the algorithmic advantage.
  2. [Abstract vs. Sections 3.5 and 5] The abstract states 'Comprehensive experiments on 12 benchmark instances and 3 GIS-derived instances' and claims reductions of 0.92% to 73.72%. The body constructs only four synthetic lattice cases (Section 3.5) and reports results only for those cases (Table 2, Fig. 8). No GIS-derived instances or 12-instance suite appears anywhere in the experimental section. This is a major discrepancy between claimed and actual evidence. Either supply the promised experiments or remove/correct the abstract statement.
  3. [§3.2–3.3, Eqs. (3) and (7)] The feasibility constraints are only specified as abstract sets X and Y, with the statement that the 'full algebraic specification' is deliberately omitted. The network topology constraints (ring/interconnected, capacity, connectivity realization, trenching-cabling consistency) are central to the problem and to the feasibility of the repair operator. Without a precise statement of these constraints, the problem is not fully defined and the implementation cannot be independently verified. This is not a presentation nicety: the claim that the proposed method finds feasible optimal solutions depends on these sets being correctly encoded and enforced.
  4. [§5.1, L-MVNS evaluation protocol] L-MVNS is pretrained on Case-0 and then 'online fine-tuned' on the first 200 iterations of the target instance before being evaluated. This means the DRL policy has already seen part of the test instance's search trajectory, so the comparison between L-MVNS and MVNS is not out-of-sample. The claim that L-MVNS 'delivers additional gains on larger instances' is therefore at risk of in-sample tuning artifact. To support generalization, report L-MVNS without online fine-tuning (pretrained only), or use a train/test instance split and show that fine-tuning on the test instance does not materially change the conclusions.
  5. [Table 2, §5.2.4] The comparisons are based on only 10 independent runs per case. Several gaps are small (e.g., Case-2 MVNS 5295.26 vs. SNS-1 5325.11, 0.68%; Case-1 L-MVNS 2489.23 vs. MVNS 2481.42, L-MVNS is actually worse but the text only notes Case-1 is an exception). No statistical tests (e.g., Wilcoxon signed-rank, Mann–Whitney U) or confidence intervals are given. Claims of 'consistently yields the best' and 'higher stability' need quantitative support, especially for differences of a few percent.
minor comments (5)
  1. [§3.5] The benchmark generation says 'with random seed 42, we generate four scales' but the specific seeds for K-means and demand draws are not documented, which hampers exact reproducibility.
  2. [§3.5] Typo: 'CNY0.5 million CNY/km' should be '0.5 million CNY/km' or 'CNY 0.5 million/km'.
  3. [§3.4] The notation for the number of feeders, arcs, and road edges is introduced as |F|, |A|, |R|, but the earlier notation uses script letters and the symbol F is not formally defined; please align the notation.
  4. [§5.2] The text says SNS-1/2/3 use a neighborhood size of 30, while MVNS/L-MVNS use 10 per operator (total 30). This is fair, but the per-operator budget difference (30 vs. 10) should be explicitly acknowledged when interpreting operator-frequency statistics in Fig. 10.
  5. [§5.1, Eq. (10)] The reward is defined only on improvement; the baseline function b(s_i) is described as a three-layer fully connected network but its inputs and training target are not specified. A short clarification would help reproducibility.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the optimization and evaluation chain is self-contained; the noted issues are validation/interpretation gaps, not definitional reductions.

full rationale

The paper's claimed derivation chain starts from a defined two-stage objective (Eqs. (2) and (6)) and an integrated formulation (Eq. (8)); it does not import that objective from the algorithm's outputs. Initialization (HGS + A*) and the MVNS operators are evaluated with the same Eq. (6) cost function, so the search is optimizing the stated criterion rather than re-labeling a fitted value. The external baselines MCWS and HGS are independent methods from the literature, and SNS-1/2/3 are ablations of the proposed operator set, so the core algorithmic comparison is not self-referential. The two concerns that deserve attention are not circularities: (i) L-MVNS is fine-tuned on the first 200 iterations of each target instance before being compared on that same instance (Section 5.1), so the DRL's reported gain over MVNS is in-sample rather than an out-of-sample prediction; (ii) the 30–50% headline compares path-aware methods with relation-only baselines that are evaluated under a different cost model ('equating trench length with cable length,' Section 5.2), so that number reflects the trench-sharing formulation and the 3:1 cost ratio more than an independent algorithmic prediction. The abstract's promised GIS-derived instances are also absent from the body, an external-validity gap. None of these make any equation equal to its own input or make a fitted parameter masquerade as a derivation.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The central claim rests mainly on benchmark modeling choices and on the completeness of omitted constraints, not on parameters fit to external data. The 30–50% saving number is sensitive to the cost ratio and the synthetic road model. No new physical entities are introduced.

free parameters (5)
  • trenching-to-cable cost ratio = c_tr = 1.5, c_ca = 0.5 M CNY/km (3:1)
    Drives the magnitude of trench-sharing savings that generate the 30–50% cost gap; adopted from [7] without validation for urban MV networks.
  • benchmark instance parameters = Q=10 MVA, demand U[2,5] MVA, C_max=6, seed 42
    Used to construct Cases 1–4; no sensitivity analysis on these values, though they affect topology feasibility and absolute costs.
  • perturbation size schedule thresholds = kappa=2/4/6/8 at stagnation thresholds 20/30/40
    Hand-set in Eq. (9); no sensitivity analysis is reported for these thresholds.
  • DRL exploration probability = 70% agent / 30% random
    Chosen to balance exploration; no sensitivity analysis is reported, and it directly affects the L-MVNS vs MVNS comparison.
  • running hyperparameters = N=10 per operator, 600 iterations, 200s initialization time
    Selected by limited sensitivity analysis on Case-3 only; values may not be optimal for all instance scales.
assumptions (4)
  • ad hoc to paper The omitted combinatorial constraints in abstract sets X (Eq. 3) and Y (Eq. 7) are correctly encoded and satisfied by HGS, A*, and MVNS repair.
    Authors deliberately omit the full algebraic constraints, so feasibility of solutions is asserted rather than proven or fully specified.
  • domain assumption Road networks can be modeled as directed graphs with cables restricted to edges and linear cost functions (Section 3.3).
    Standard GIS abstraction; ignores construction details such as jointing, voltage drops, and digging width.
  • ad hoc to paper The DRL policy pretrained on Case-0 and fine-tuned for 200 iterations on the target instance transfers usefully to larger instances (Section 5.1).
    No out-of-sample evaluation; transfer evidence is indirect and the gains are small.
  • domain assumption Synthetic lattice instances approximate real urban layouts (Section 3.5).
    Regular 1-km grids with K-means substations are used despite the abstract promising real GIS-derived instances.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Learning to Sample in Variable Neighborhood Search Algorithm for Urban Cable Routing Optimization." pith.science (2026). https://pith.science/paper/DY6VVHYA

@misc{pith2026251219321,
  author       = {Pith},
  title        = {Pith review of: Learning to Sample in Variable Neighborhood Search Algorithm for Urban Cable Routing Optimization},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DY6VVHYA}},
  note         = {Machine review of arXiv:2512.19321}
}
read the original abstract

Urban underground cable construction is essential for enhancing power grid reliability, yet the high construction costs demand systematic optimization. Constrained by road network infrastructure, this problem requires consideration of both connectivity relationships between substations and specific routing strategies along road networks, constituting a large-scale bilevel combinatorial optimization problem. Insufficient attention to routing subproblems in traditional research and simplistic algorithmic designs ill-equipped for large-scale optimization leave substantial room for advancement. To navigate the enormous combinatorial search space, we propose a learning-assisted variable neighborhood search (L-VNS) algorithm integrating four key components. First, an auxiliary task focusing on the upper-level connectivity subproblem generates high-quality initial solutions by employing hybrid genetic search for connection optimization and A* for detailed path routing. Subsequently, the algorithm iteratively refines the connectivity topology using variable neighborhood search with three complementary operators. A multi-agent deep reinforcement learning module adaptively guides probabilistic neighborhood sampling by jointly encoding upper-level connectivity patterns and lower-level routing structures, effectively exploiting problem structure. Finally, a modified A* operator re-plans lower-level paths affected by neighborhood modifications to ensure feasibility and solution completeness. Comprehensive experiments on 12 benchmark instances and 3 GIS-derived instances demonstrate the superiority of L-VNS, achieving total construction cost reductions of 0.92% to 73.72% compared to representative approaches. Ablation studies and sensitivity analyses further validate the effectiveness and robustness of the proposed algorithm.

Figures

Figures reproduced from arXiv: 2512.19321 by the authors.

Figure 1
Figure 1. Typical topologies of urban MV distribution cable networks: (a) Interconnected network: MV substations are linked by feeders that originate from one HV substation and terminate at another; (b) Ring network: MV substations are linked by feeders that both originate from and terminate at the same HV substation. is to determine the lowest-cost cable paths between sub￾stations, while satisfying capacity constraints, conn… view at source ↗
Figure 2
Figure 2. Schematic diagrams of the urban cable routing problem underscore two key points. First, as shown by panel (a) versus panel (b), cables cannot be laid along arbitrary straight￾line segments; rather, their paths must adhere strictly to the existing road network topology. Second, as illustrated by panel (c) versus panel (d), planning multiple feeders in parallel along the same road segments can exploit spatial overlap … view at source ↗
Figure 3
Figure 3. visualizes them. • Case-1: 20 × 20 grid, 𝑁𝑚𝑣 = 30, 𝑁ℎ𝑣 = 5; • Case-2: 30 × 30 grid, 𝑁𝑚𝑣 = 50, 𝑁ℎ𝑣 = 6; • Case-3: 30 × 30 grid, 𝑁𝑚𝑣 = 80, 𝑁ℎ𝑣 = 7; • Case-4: 30 × 30 grid, 𝑁𝑚𝑣 = 100, 𝑁ℎ𝑣 = 8. (a) Case-1 (b) Case-2 (c) Case-3 (d) Case-4 [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: The proposed L-MVNS algorithm comprises three modules: solution initialization, learning-assisted neighborhood generation, and multi-operator variable neighborhood search. The final solution is iteratively refined in the latter two modules. Concretely, at iteration 𝑡, …
Figure 5
Figure 5. Figure 5: The operation diagrams for the three destruction operators: squares indicate HV substations, circles denote MV substations, orange and blue modules signify two independent feeders, gray dotted lines represent removed segments, and orange and blue dotted lines represent…
Figure 6
Figure 6. Figure 6: Schematic of the neural network architecture and feedforward computation. Each agent comprises a sequential stack of an LSTM layer, a multi-head attention layer, and a fully connected layer. The network ingests a three-dimensional tensor that encodes the two-dimensiona…
Figure 7
Figure 7. Figure 7: reports the reward curves for the multi-agent aggregation model and each of the three individual agents during training. The reward, defined in Equation (10), mea￾sures the proportion of the total cost reduced by a single neighborhood-search iteration. Across runs, the…
Figure 8
Figure 8. Figure 8: Boxplots of solution costs for all methods on Cases 1–4 over 10 independent runs. Boxes show the interquartile range with median lines; whiskers indicate non-outlier ranges; red points mark the mean and are connected across methods. Lower values indicate better perform…
Figure 9
Figure 9. Figure 9: Representative solutions for Cases 1–4 obtained by HGS (top row) and L-MVNS (bottom row). L-MVNS substantially adjusts both substation connectivity and route choices, with parallel cable placement along shared trenches (darker lines) reducing total cost. and specific r…
Figure 12
Figure 12. Figure 12: Boxplots of MVNS performance on Case-3 over 10 runs with neighborhood size per operator from 5 to 25. Red dots (mean values) follow an approximate decaying-exponential trend. performance on Case-3 across 10 independent runs for each setting. The average cost drops mar…
Figure 11
Figure 11. Figure 11: Boxplots of MVNS performance on Case-3 over 10 runs with initialization time from 10 to 600 seconds. Red dots (mean values) follow an approximate decaying-exponential trend. marginal reduction in total cost plateaus. The mean perfor￾mance at each initialization time (…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

42 extracted references

  1. [1]

    Low-carbon urban–rural modern energy systems with en- ergy resilience under climate change and extreme events in china—a state-of-the-art review

    Zhou, Y.. Low-carbon urban–rural modern energy systems with en- ergy resilience under climate change and extreme events in china—a state-of-the-art review. Energy and Buildings 2024;:114661

    2024

  2. [2]

    A multi-period multi-criteria replacement and rejuvenation planning of underground cables in urban distribution networks

    Milošević, N.D., Popović, Ž.N., Kovački, N.V.. A multi-period multi-criteria replacement and rejuvenation planning of underground cables in urban distribution networks. International Journal of Elec- trical Power & Energy Systems 2023;149:109018

    2023

  3. [3]

    Applicationofevolutionary algorithmsfortheplanningofurbandistributionnetworksofmedium voltage

    Diaz-Dorado, E.,Cidrás, J.,Míguez, E.. Applicationofevolutionary algorithmsfortheplanningofurbandistributionnetworksofmedium voltage. IEEE Transactions on Power Systems 2002;17(3):879–884

    2002

  4. [4]

    Optimal wind farm collector system topol- ogy design considering total trenching length

    Dutta, S., Overbye, T.J.. Optimal wind farm collector system topol- ogy design considering total trenching length. IEEE Transactions on Sustainable Energy 2012;3(3):339–348

    2012

  5. [5]

    Heuristicalgorithmsforthe windfarmcableroutingproblem

    Cazzaro, D.,Fischetti, M.,Fischetti, M.. Heuristicalgorithmsforthe windfarmcableroutingproblem. AppliedEnergy2020;278:115617

  6. [6]

    Hazard identification and analysis of urban utility tunnels in china

    Bai, Y., Zhou, R., Wu, J.. Hazard identification and analysis of urban utility tunnels in china. Tunnelling and Underground Space Technology 2020;106:103584

    2020

  7. [7]

    Electrical layout optimization of onshore wind farms based on a two-stage approach

    Deveci, K., Barutçu, B., Alpman, E., Taşcıkaraoğlu, A., Erdinç, O.. Electrical layout optimization of onshore wind farms based on a two-stage approach. IEEE Transactions on Sustainable Energy 2019;11(4):2407–2416

    2019

  8. [8]

    Combined layout and cable optimizationofoffshorewindfarms.EuropeanJournalofOperational Research 2023;311(1):301–315

    Cazzaro, D., Koza, D.F., Pisinger, D.. Combined layout and cable optimizationofoffshorewindfarms.EuropeanJournalofOperational Research 2023;311(1):301–315

    2023

Show all 42 references
  1. [9]

    Optimization of cable layout designs for large offshore wind farms

    Ulku, I., Alabas-Uslu, C.. Optimization of cable layout designs for large offshore wind farms. International Journal of Energy Research 2020;44(8):6297–6312

    2020

  2. [10]

    Adesignmethod- ology for wind farm layout considering cable routing and economic benefitbasedongeneticalgorithmandgeosteiner

    Wu, Y.,Zhang, S.,Wang, R.,Wang, Y.,Feng, X.. Adesignmethod- ology for wind farm layout considering cable routing and economic benefitbasedongeneticalgorithmandgeosteiner. RenewableEnergy 2020;146:687–698

    2020

  3. [11]

    Optimal cabling of deep-sea far-shore wind farm considering complex seabed and restricted zones

    Tao, S., Jiang, F., Yang, J., Feijóo-Lorenzo, A.E.. Optimal cabling of deep-sea far-shore wind farm considering complex seabed and restricted zones. IEEE Transactions on Industrial Informatics 2025;21(7):5557–5568

    2025

  4. [12]

    Optimal feeder routing in urban distribution networks planning with layout constraints and losses

    Bosisio, A.,Berizzi, A.,Amaldi, E.,Bovo, C.,Sun, X.A.. Optimal feeder routing in urban distribution networks planning with layout constraints and losses. Journal of Modern Power Systems and Clean Energy 2020;8(5):1005–1014

    2020

  5. [13]

    Gis-based urban distribution networks planning with 2-step ladder topology considering electric power cable joints

    Bosisio, A.,Berizzi, A.,Bovo, C.,Amaldi, E.,Fratti, S.. Gis-based urban distribution networks planning with 2-step ladder topology considering electric power cable joints. In: 2018 AEIT International Annual Conference. IEEE; 2018, p. 1–6

    2018

  6. [14]

    A gis-based approach for high-level distribution networks expansionplanninginnormalandcontingencyoperationconsidering reliability

    Bosisio, A.,Berizzi, A.,Amaldi, E.,Bovo, C.,Morotti, A.,Greco, B., et al. A gis-based approach for high-level distribution networks expansionplanninginnormalandcontingencyoperationconsidering reliability. Electric Power Systems Research 2021;190:106684

    2021

  7. [15]

    Apracticalurbandistribution networkplanningmethodwithgeographicinformationsystem

    Wang, Z.,Shen, X.,Sun, H.,Wu, Q.. Apracticalurbandistribution networkplanningmethodwithgeographicinformationsystem. IEEE Transactions on Power Systems 2024

    2024

  8. [16]

    An milp model for urban distribution network planning considering street layout and block loads

    Li, Z., Wu, W., Zhang, B., Tai, X.. An milp model for urban distribution network planning considering street layout and block loads. In: 2019 IEEE Power & Energy Society General Meeting (PESGM). IEEE; 2019, p. 1–5

    2019

  9. [17]

    Substation secondary cable path optimization design based on 3d simulation and improved dijkstra algorithm

    Li, S., Zhang, Y., Ling, Z., Zhang, X.. Substation secondary cable path optimization design based on 3d simulation and improved dijkstra algorithm. In: Journal of Physics: Conference Series; vol

  10. [18]

    Optimal routing and design of radial medium voltage power distribution networks considering cables short cir- cuit withstand-capacity

    Youssef, K.H.. Optimal routing and design of radial medium voltage power distribution networks considering cables short cir- cuit withstand-capacity. IEEE Transactions on Power Delivery 2022;37(4):2429–2439

    2022

  11. [19]

    Genetic algorithm and gis enhancedlongtermplanningoflargelinkstructureddistributionsys- tems.In:LESCOPE’02.2002LargeEngineeringSystemsConference on Power Engineering

    Skok, M., Skrlec, D., Krajcar, S.. Genetic algorithm and gis enhancedlongtermplanningoflargelinkstructureddistributionsys- tems.In:LESCOPE’02.2002LargeEngineeringSystemsConference on Power Engineering. Conference Proceedings. IEEE; 2002, p. 55– 60

    2002

  12. [20]

    A genetic algorithm for finding microgrid cable layouts

    Göttlicher, M., Wolf, M.. A genetic algorithm for finding microgrid cable layouts. In: Proceedings of the Thirteenth ACM International Conference on Future Energy Systems. 2022, p. 1–16

    2022

  13. [21]

    A route optimization method for urban powerpipelinenetworklayout

    Liu, R., Li, J., Chen, Q.. A route optimization method for urban powerpipelinenetworklayout. In:20165thInternationalConference on Computer Science and Network Technology (ICCSNT). IEEE; 2016, p. 381–385

    2016

  14. [22]

    Multi-objective optimization for submarine optical cable route planning based on cross reinforcement learning

    Zhao, Z., Gao, G., Gan, W., Zhang, J., Wang, Z., Wang, H., et al. Multi-objective optimization for submarine optical cable route planning based on cross reinforcement learning. Journal of Optical Communications and Networking 2024;16(10):1018–1033

    2024

  15. [23]

    Multistagecableroutingthroughhierarchicalimitationlearning

    Luo, J., Xu, C., Geng, X., Feng, G., Fang, K., Tan, L., etal. Multistagecableroutingthroughhierarchicalimitationlearning. IEEE Transactions on Robotics 2024;40:1476–1491

    2024

  16. [24]

    Knowledge classification-assisted evolutionary multitasking for two- task multiobjective optimization problems

    Wang, X., Kang, Q., Zhou, M., Deng, Q., Fan, Z., Liu, H.. Knowledge classification-assisted evolutionary multitasking for two- task multiobjective optimization problems. IEEE/CAA Journal of Automatica Sinica 2025;12(6):1176–1193

    2025

  17. [25]

    Machine learning assisted multiobjective evolutionary algorithm for routing andpacking

    Liu, F., Zhang, Q., Zhu, Q., Tong, X., Yuan, M.. Machine learning assisted multiobjective evolutionary algorithm for routing andpacking. IEEETransactionsonEvolutionaryComputation2024

  18. [26]

    Hybridizationofevolutionaryalgorithmanddeepreinforcement learningformultiobjectiveorienteeringoptimization

    Liu, W., Wang, R., Zhang, T., Li, K., Li, W., Ishibuchi, H., etal. Hybridizationofevolutionaryalgorithmanddeepreinforcement learningformultiobjectiveorienteeringoptimization. IEEETransac- tions on Evolutionary Computation 2023;27(5):1260–1274

    2023

  19. [27]

    Computers & Operations Research 2022;140:105643

    Vidal, T..Hybridgeneticsearchforthecvrp:Open-sourceimplemen- tation and swap* neighborhood. Computers & Operations Research 2022;140:105643

    2022

  20. [28]

    IEEEtransactionsonSystems Science and Cybernetics 1968;4(2):100–107

    Hart, P.E.,Nilsson, N.J.,Raphael, B..Aformalbasisfortheheuristic determinationofminimumcostpaths. IEEEtransactionsonSystems Science and Cybernetics 1968;4(2):100–107

    1968

  21. [29]

    Variable-depth large neighborhood search algorithm for cable routing in distributed photovoltaic systems

    Qiu, A., Yang, Z.. Variable-depth large neighborhood search algorithm for cable routing in distributed photovoltaic systems. Au- tomation in Construction 2024;168:105839

    2024

  22. [30]

    Gis spatial analysis applied to electric line routing optimization

    Monteiro, C., Ramírez-Rosado, I.J., Miranda, V., Zorzano- Santamaría, P.J.,García-Garrido, E.,Fernández-Jiménez, L.A.. Gis spatial analysis applied to electric line routing optimization. IEEE transactions on Power Delivery 2005;20(2):934–942

    2005

  23. [31]

    Optimizingpowercableroutingusingai

    Larsson, J.. Optimizingpowercableroutingusingai. Master’sthesis; Chalmers University of Technology; Gothenburg, Sweden; 2025

    2025

  24. [32]

    Electric Power Systems Research 2022;211:108217

    Trageser, M., Pape, M., Frings, K., Erlinghagen, P., Kurth, M., Vertgewall, C.M.,etal.Automatedroutingoffeedersinelectricaldis- tribution grids. Electric Power Systems Research 2022;211:108217

    2022

  25. [33]

    Optimal routing an ungrounded electrical distribution system based on heuristic method with micro grids integration

    Pavón, W., Inga, E., Simani, S.. Optimal routing an ungrounded electrical distribution system based on heuristic method with micro grids integration. Sustainability 2019;11(6):1607

    2019

  26. [34]

    Planning of a resilient undergrounddistributionnetworkusinggeoreferenceddata

    Valenzuela, A., Inga, E., Simani, S.. Planning of a resilient undergrounddistributionnetworkusinggeoreferenceddata. Energies 2019;12(4):644

    2019

  27. [35]

    A practical large-scale distribution network planning model based on elite ant-q

    Wang, Z., Lin, D., Zeng, G., Yu, T.. A practical large-scale distribution network planning model based on elite ant-q. IEEE Access 2020;8:58912–58922

    2020

  28. [36]

    Integrated reliability centered distribution system planning—cable routing and switch placement

    Duvnjak Zarkovic, S., Shayesteh, E., Hilber, P.. Integrated reliability centered distribution system planning—cable routing and switch placement. Energy Reports 2021;7:3099–3115

    2021

  29. [37]

    Transmission and distribution electrical engineering

    Bayliss, C., Hardy, B.. Transmission and distribution electrical engineering. Elsevier; 2012

    2012

  30. [38]

    Gu, D., Dai, H., Zeng, J.. Full-cable medium voltage distribution networkplanningbasedonincrementalshortestpathmethod.Electric First Author et al.:Preprint submitted to Elsevier Page 13 of 14 Learning-Assisted Multi-Operator Variable Neighborhood Search for Urban Cable Routing...

    2019

  31. [39]

    PyVRP:ahigh-performanceVRP solver package

    Wouda, N.A.,Lan, L.,Kool, W.. PyVRP:ahigh-performanceVRP solver package. INFORMS Journal on Computing 2024;36(4):943– 955

    2024

  32. [40]

    Longshort-termmemory

    Graves, A.. Longshort-termmemory. Supervisedsequencelabelling with recurrent neural networks 2012;:37–45

    2012

  33. [41]

    Attention is all you need

    Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A.N., et al. Attention is all you need. Advances in neural information processing systems 2017;30. First Author et al.:Preprint submitted to Elsevier Page 14 of 14

    2017

  34. [2136]

    IOP Publishing; 2021, p. 012009

    2021

Pith tools

Reviewed August 3, 2026 · model on record in the stance chip above.