logistics-inventory-forecasting-operations-and-solvers
Use when the surrounding system matters as much as the algorithm: inventory policy and safety stock, forecasting and its error measures, the vendor and product landscape, warehouse operations including slotting and picking, last-mile delivery, the data layer that decides most project outcomes, dynamic and stochastic problems including re-optimization, and the solver and tooling options.
- License 许可
- MIT
- 最近更新
- 2026-11-15
- Stars
- ★ 5
- Forks
- 3
01 前言 Front Matter
- name
- logistics-inventory-forecasting-operations-and-solvers
- description
- Use when the surrounding system matters as much as the algorithm: inventory policy and safety stock, forecasting and its error measures, the vendor and product landscape, warehouse operations including slotting and picking, last-mile delivery, the data layer that decides most project outcomes, dynamic and stochastic problems including re-optimization, and the solver and tooling options.
03 SKILL.md 正文
Logistics and Optimization Software: Inventory, Forecasting, Warehouse and Last-Mile Operations, Data, Stochastic Problems, and Solvers
Part 4 of 5 of the Logistics and Optimization Software reference (plugin
logistics-software-optimization), covering §17–§24. Sibling skills:logistics-why-projects-fail-complexity-and-modeling(§0–§5),logistics-constraint-programming-metaheuristics-and-bounds(§6–§9),logistics-routing-packing-scheduling-and-network-design(§10–§16),logistics-reference(§25–§31). Section numbers are shared across the set; a reference written as §N →skillpoints into that sibling skill.Currency: The algorithms and complexity results are permanent. Two areas are live. See §25 →
logistics-referencefor the solver landscape and the quantum and AI optimization claims.
⚠️ Scope. For engineers building or integrating logistics systems. Complements a data-engineering reference (pipelines), an operations reference (process), and a business reference (unit economics).
⚠️ GOTCHA boxes mark things that sink projects or silently produce wrong answers.
The three ideas that organize this document:
- ⚠️ Optimization projects almost never fail on the algorithm. They fail on data quality, on constraints nobody told you about, and on solutions the drivers refuse to run (§1 →
logistics-why-projects-fail-complexity-and-modeling). The maths is the easy part and it is not where your time goes.- ⚠️ NP-hard does not mean unsolvable — it means no guaranteed-fast exact method. Real instances with tens of thousands of stops are solved to within a few percent of optimal every night by well-engineered heuristics (§2 →
logistics-why-projects-fail-complexity-and-modeling, §8 →logistics-constraint-programming-metaheuristics-and-bounds).- ⚠️ A 2% better solution that operations won't execute is worth 0%. Stability, explainability and consistency with yesterday's plan routinely matter more than the objective value (§1 →
logistics-why-projects-fail-complexity-and-modeling, §26 →logistics-reference).
§17. Inventory
EOQ ⚠️ the classic square-root formula. Assumptions rarely hold,
and it's still a useful order-of-magnitude sanity check
⚠️ NEWSVENDOR single period, stochastic demand. Optimal service level =
Cu / (Cu + Co) — underage cost over total. ⚠️ THE key insight: your
service level should follow from the COST ASYMMETRY, not from a target
somebody picked
(s,S) / (R,Q) reorder point policies
SAFETY STOCK ⚠️ ≈ z · σ_demand-over-lead-time. Note it scales with the
square root of lead time — ⚠️ so REDUCING LEAD TIME VARIABILITY beats
reducing average lead time, usually by a lot
MULTI-ECHELON ⚠️ where to hold stock across a network. Risk pooling:
consolidated inventory needs less safety stock for the same service
level, scaling with √n
ABC / XYZ segmentation by value and by variability
⚠️ GOTCHA — the bullwhip effect is a structural property, not a forecasting failure. ⚠️ Demand variability amplifies up the supply chain because of order batching, lead time, price promotions and rationing behaviour — it arises from the STRUCTURE even with rational actors. You mitigate it with information sharing and shorter lead times, not with better local forecasts.
§18. Forecasting
⚠️ Keep expectations calibrated: for intermittent, lumpy SKU-level demand, sophisticated methods frequently fail to beat simple ones. Methods: moving average, exponential smoothing / Holt-Winters, ARIMA, Croston's method (⚠️ specifically for intermittent demand), gradient boosting on engineered features, and hierarchical reconciliation (⚠️ forecast at multiple levels and reconcile so they sum consistently). ⚠️ The M-competitions are the honest benchmark literature here, and the recurring finding is that simple methods and combinations are extremely hard to beat, ⚠️ with ML methods winning mainly where there is cross-series structure to exploit. ⚠️ Forecast accuracy metrics: MAPE (⚠️ breaks on zeros and is asymmetric — a real problem for intermittent demand), MASE (⚠️ better), RMSE, bias. ⚠️ Track BIAS separately from accuracy — a consistently biased forecast is a systematically mis-stocked warehouse, and it's invisible in symmetric error metrics.
PART III — SYSTEMS
§19. The Landscape
ERP the system of record — orders, inventory, finance
OMS order management, allocation, sourcing decisions
⚠️ WMS warehouse management — receiving, putaway, picking, packing, shipping
⚠️ TMS transportation management — planning, tendering, execution, freight audit
YMS yard management — trailers, docks, gates
WES/WCS execution/control — the real-time layer talking to conveyors,
sorters and robots
LMD last-mile delivery platforms, driver apps, customer tracking
⚠️ The integration reality: these are usually different vendors with different data models, and the join keys don't line up. ⚠️ EDI (X12/EDIFACT) remains pervasive in freight — 214 shipment status, 204 load tender, 990 response, 210 invoice — and API-first carriers coexist with partners who will only do EDI over an SFTP drop. ⚠️ Budget for the integration layer as a first-class component; in most logistics projects it is larger than the optimization component.
§20. Warehouse Operations
⚠️ Travel time dominates manual picking — commonly cited as roughly half of picker time — so most warehouse optimization is really travel reduction.
SLOTTING ⚠️ put fast movers near dispatch; account for ergonomics
(golden zone) and for affinity (items ordered together)
PICK PATH ⚠️ TSP on the warehouse graph. S-shape and return heuristics
are common and near-optimal for simple layouts
BATCHING ⚠️ pick multiple orders per trip — usually the single biggest
travel win available
ZONING pick-and-pass vs pick-and-sort
WAVE vs WAVELESS ⚠️ batched release vs continuous flow
GOODS-TO-PERSON ⚠️ AS/RS, shuttles, AMRs — inverts the problem: now you
optimize robot fleet movement and storage assignment instead
⚠️ Labour standards and ergonomics are constraints, not preferences — and an optimizer that maximizes picks per hour without them produces injury rates and turnover that cost more than the throughput gained.
§21. Last Mile
⚠️ The most expensive segment — commonly cited as a large share of total delivery cost — and the most constrained. Key issues: failed first delivery attempts (⚠️ each retry is close to a full additional delivery cost, which is why access codes, safe-place instructions and notification quality have outsized ROI), time window promises vs cost (⚠️ narrow windows are extremely expensive; letting customers self-select from optimizer-suggested windows is far cheaper than accepting arbitrary ones), parking and access reality that no routing model captures, gig vs employed fleets, lockers and pickup points, and returns as a first-class flow rather than an exception. ⚠️ Service time is where routing plans die. ⚠️ Modeling every stop at a flat 3 minutes when apartment buildings take 12 produces a plan that is an hour behind by midday. Measure actual service times per stop type and per location, and feed the distribution back (§22).
§22. ⚠️ Data
⚠️ Where the projects actually live or die.
⚠️ ADDRESSES unstructured, misspelled, ambiguous, missing units.
⚠️ Geocoding returns a point that may be a street centroid, a rooftop,
or the middle of a postcode — and the QUALITY FLAG matters more than
the coordinate. Rooftop vs interpolated is the difference between
arriving and circling
⚠️ ROAD NETWORKS OSM, HERE, TomTom. Turn restrictions, one-ways, height
and weight limits, seasonal closures
⚠️ DISTANCE MATRIX n² entries. ⚠️ THE dominant cost and latency issue at scale
TRAVEL TIME ⚠️ time-dependent, and asymmetric — A→B ≠ B→A
⚠️ SERVICE TIME measure it; don't assume it (§21)
MASTER DATA ⚠️ vehicle capacities, driver skills, customer constraints.
Usually stale, and nobody owns it
⚠️ GOTCHA — the distance matrix is the hidden scaling wall. ⚠️ 1,000 stops = a million matrix entries; 10,000 stops = 100 million. Commercial matrix APIs charge per element and rate-limit, so a naive nightly full-matrix rebuild is both slow and expensive. ⚠️ The mitigations: self-host a routing engine (OSRM, Valhalla, GraphHopper) — which turns a per-element cost into a fixed server cost; compute only the k-nearest neighbours per stop, since a good solution never uses most of the matrix; cache aggressively — road networks change slowly; and ⚠️ use haversine distance for early filtering and clustering, never for the final plan.
⚠️ And a warning about the cheap shortcut: straight-line distance times a fudge factor is fine for clustering and wrong for planning. ⚠️ The error is not uniform — it's worst exactly where geography constrains routing (rivers, motorways, one-way systems), which is precisely where your plan will fail.
§23. Dynamic and Stochastic
⚠️ Real logistics is not a static problem solved once.
DYNAMIC ⚠️ orders arrive during execution; vehicles are en route
STOCHASTIC ⚠️ travel times, service times and demand are random variables
ONLINE ⚠️ decide without knowing the future. Competitive ratio analysis
ROLLING HORIZON ⚠️ re-optimize periodically over a moving window. THE
standard practical approach
Techniques: rolling-horizon re-optimization; scenario-based stochastic programming; sample average approximation; robust optimization (⚠️ optimize the worst case — often too conservative for logistics); chance constraints; and reinforcement learning (⚠️ genuinely promising for dispatch-style sequential decisions, and much harder to validate and deploy than the literature implies). ⚠️ The practical patterns that matter more than the theory:
- ⚠️ Buffer time in the plan. A plan with zero slack fails on the first delay and cascades.
- ⚠️ Re-optimize on a schedule AND on trigger events, not continuously — continuous
re-optimization destabilizes the plan (§1 →
logistics-why-projects-fail-complexity-and-modeling). - ⚠️ Freeze a near horizon. Don't change what a driver is doing in the next 30 minutes. This is both operationally necessary and a large search-space reduction.
- Anticipate rather than react where you can — positioning idle capacity toward expected demand.
§24. Solvers and Tooling
COMMERCIAL MIP ⚠️ Gurobi · IBM CPLEX · FICO Xpress · COPT (§25.1)
OPEN SOURCE MIP ⚠️ HiGHS (the current default) · SCIP (excellent, check
licence terms) · CBC (older, slower)
CP ⚠️ OR-Tools CP-SAT (outstanding, free) · IBM CP Optimizer
ROUTING ⚠️ OR-Tools routing library · VROOM (open source, fast) ·
jsprit · commercial routing engines
MODELING LAYERS ⚠️ Pyomo · PuLP · JuMP (Julia) · AMPL/GAMS · linopy
ROAD ROUTING ⚠️ OSRM · Valhalla · GraphHopper (self-host; §22)
FIRST-ORDER/GPU ⚠️ PDLP and cuPDLP — GPU LP for very large instances (§25.1)
⚠️ Sensible defaults for a team starting out: OR-Tools for routing and scheduling
(free, well-documented, genuinely good); HiGHS for LP/MIP; a self-hosted OSRM for
distances; Pyomo or JuMP if you want solver portability. ⚠️ Buy a commercial MIP
licence when you have demonstrated that solve time is your binding constraint — and not
before, because §1 → logistics-why-projects-fail-complexity-and-modeling says it usually isn't.
04 同仓库更多 Skills
white-paper-writing
Use when writing white papers, technical reports, policy documents, thought leadership pieces, or any long-form authoritative document. Triggers on requests to write, structure, outline, or review white papers, technical reports, position papers, policy briefs, or evidence-driven arguments. Also use for B2B content marketing strategy, technical documentation requiring authority, academic-style reports, or any document that must educate while persuading. Use when someone asks "how should I structure this" for a long-form document.
prose-technical-mechanics
Use when editing or writing prose at a technical/mechanical level: choosing between Anglo-Saxon and Latinate diction, placing information in topic vs stress positions, writing periodic vs cumulative sentences, handling sentence rhythm, punctuation as logic, transitions, paragraph structure, or phonesthetics. Use when someone wants to understand WHY prose works or doesn't, not just surface-level style advice. Triggers on word choice, sentence structure, emphasis, rhythm, diction, or any deep writing mechanics question.
prose-mechanics
Use when writing, editing, or analyzing prose style; when trying to avoid generic AI-sounding writing; when applying specific prose styles such as journalistic, academic, narrative, KJV biblical, Hemingway plain, Latinate/Jamesian, Orwellian, Augustan, Victorian, Early Modern (Bacon, Browne, Milton), or free indirect discourse; or when coaching someone on writing voice and mechanics. Triggers on any request to improve, analyze, or write in a specific prose style, to fix "AI-sounding" writing, to make prose more vivid or concrete, to vary sentence rhythm, or to understand how a particular writer achieves their effects.
narrative-structure
Use when structuring stories, pitches, presentations, white papers, or any persuasive narrative; analyzing the hero's journey in fiction or non-fiction; applying StoryBrand for business communication; designing narrative arcs for thought leadership; or understanding Vogler, Snyder, Propp, or Campbell frameworks. Triggers on any narrative structure, storytelling, pitch structure, StoryBrand, hero's journey, three-act structure, Freytag, Save the Cat, narrative arc, or "how should I structure this story/pitch/presentation" request.
legal-brief-writing
Use when writing legal briefs, motions, or any court filing; when analyzing judicial cognition and persuasion; when applying elite brief-writing techniques from Roberts, Clement, Kagan, Garner; or when structuring legal arguments for maximum persuasion. Also use for autistic/neurodivergent advocates who need to understand how their natural communication style maps to elite legal advocacy. Triggers on any legal writing, brief drafting, motion preparation, or legal argument structuring request.
journal-publishability-criteria
Use whenever a research paper, article or preprint is being written, revised, reviewed or published, and specifically before every publication of docs/paper.md. Covers what legitimate journals actually gate on, in order of gating — the editor's desk screen (scope, novelty, package problems), what reviewers judge (methodological soundness, claims the evidence can carry, statistical and qualitative reporting), EQUATOR reporting guidelines, the ICMJE authorship criteria, COPE's view of text recycling, the near-universal generative-AI policy (no AI author, disclosure, author responsibility), TOP 2025 transparency standards, how to verify a journal is legitimate, submitting without an institution, and a pre-submission checklist. Triggers on requests to evaluate, review, prepare or submit a paper, to judge publishability, to write a declarations or data-availability section, to disclose AI use, or to check a journal.