InfiniSynapse
What is inventory forecasting?什么是库存预测？
Inventory forecasting estimates future stock positions by combining expected demand, current inventory, scheduled supply, replenishment timing, and inventory policy. Its useful outputs are a time-phased view of projected on-hand stock, shortage or excess risk, and the quantity and timing of potential net requirements.
库存预测通过组合预期需求、当前库存、已排程供应、补货时间和库存政策，估算未来库存位置。有用的输出包括按时间展开的预计在手库存、短缺或过剩风险，以及潜在净需求的数量与时间。
The word “forecast” can cause confusion. A demand forecast predicts what customers or dependent users may require. An inventory forecast takes that demand input and performs a stock-flow projection: beginning inventory plus usable receipts minus demand and other issues. It may then compare the projection with safety stock, minimum stock, capacity, shelf-life, or service constraints.
“预测”一词容易引起混淆。需求预测估计客户或相关使用方可能需要多少；库存预测则把需求作为输入进行库存流量推演：期初库存加可用收货，减去需求和其他出库，再与安全库存、最低库存、容量、保质期或服务约束比较。
A forecast is a decision aid, not a promise. It depends on transaction accuracy, receipt reliability, the quality of demand estimates, the chosen time bucket, and explicit assumptions about what supply can arrive within the horizon. A single total for the month is rarely enough when an item can run out in week one and recover in week four.
预测是决策辅助，而不是承诺。它依赖交易准确性、收货可靠性、需求估计质量、时间桶选择，以及对预测期内哪些供应能够到达的明确假设。如果物料可能在第一周缺货、第四周恢复，只看整月总量通常不够。
Inventory forecasting vs demand forecasting and replenishment库存预测、需求预测与补货的区别
| Discipline领域 | Primary question核心问题 | Typical output典型输出 | Boundary边界 |
|---|---|---|---|
| Demand forecasting需求预测 | How much demand is expected?预计需求是多少？ | Baseline forecast, range, bias, and error基线预测、区间、偏差和误差 | Does not know current stock or confirmed receipts by itself本身并不知道当前库存或确认收货 |
| Inventory forecasting库存预测 | How will stock change over time?库存将如何随时间变化？ | Projected inventory, risk dates, and net requirements预计库存、风险日期和净需求 | Projection may be unconstrained unless supply feasibility is modeled若未建模供应可行性，推演可能不受约束 |
| Inventory planning库存计划 | What policy and targets should govern stock?什么政策与目标应管理库存？ | Service targets, buffers, order rules, and scenarios服务目标、缓冲、订货规则和场景 | Requires economic and operational tradeoffs需要权衡经济与运营因素 |
| Replenishment execution补货执行 | What approved order should be released?应释放什么已批准订单？ | Purchase, transfer, or production order采购、调拨或生产订单 | Needs workflow, constraints, approval, and transaction control需要工作流、约束、审批和交易控制 |
Keep the planned demand-forecasting page separate because its search intent focuses on statistical and judgmental prediction methods. This page uses the demand signal but answers a downstream operational question: whether available and expected inventory covers demand at the correct dates.
计划中的需求预测页面应保持独立，因为其搜索意图聚焦统计与判断预测方法。本页使用需求信号，但回答下游运营问题：现有及预计库存能否在正确日期覆盖需求。
Core inventory forecasting formulas库存预测核心公式
This stock-flow identity is the foundation of a time-phased inventory forecast. “Usable” matters: a purchase order expected after the need date, a receipt blocked in quality inspection, or supply already allocated elsewhere should not be treated as freely available. Demand may include customer orders, independent forecast demand, dependent production demand, transfers, reservations, scrap, or other issues according to the defined scope.
这一库存流量恒等式是分期库存预测的基础。“可用”非常重要：需求日期之后才到达的采购订单、处于质检冻结的收货或已分配给其他用途的供应，不应视为可自由使用。根据定义范围，需求可能包括客户订单、独立预测需求、相关生产需求、调拨、预留、报废或其他出库。
For continuous replenishment, the planned safety stock formula guide explains uncertainty buffers, while reorder point commonly adds expected lead-time demand to safety stock. For periodic review, protection may extend through the review interval plus lead time. Do not mix monthly demand with lead time in days without conversion.
对于连续补货，计划上线的安全库存公式指南解释不确定性缓冲，而再订货点通常把提前期预期需求与安全库存相加。定期检查的保护期可能覆盖检查间隔加提前期。不能在不换算的情况下把月需求与按天计的提前期混用。
Prepare reliable inventory forecasting data准备可靠的库存预测数据
Build the forecast at the grain where decisions occur—usually SKU-location-time bucket, sometimes with lot, status, channel, or customer allocation. Aggregate corporate inventory can look healthy while a specific location is short. Conversely, excessive detail creates sparse history and unstable models. Define both calculation grain and the hierarchy used for review.
应在实际决策粒度建立预测，通常是 SKU-地点-时间桶，有时还包括批次、状态、渠道或客户分配。公司总体库存可能健康，但某个地点仍短缺；反之，粒度过细会产生稀疏历史和不稳定模型。必须同时定义计算粒度和复核层级。
Requested demand, fulfilled sales, stockout flags, forecast snapshots, orders, promotions, returns, cancellations, substitutions, and dependent demand.
请求需求、已履约销量、缺货标记、预测快照、订单、促销、退货、取消、替代和相关需求。
On hand by status and location, reservations, allocations, quarantine, in-transit stock, backorders, expiry, units of measure, and inventory-adjustment history.
按状态与地点的在手库存、预留、分配、冻结、在途、延期交货、效期、计量单位和库存调整历史。
Open purchase, transfer, and production orders; requested, promised, shipped, and receipt dates; partial receipts; supplier calendars; minimums and pack sizes.
未结采购、调拨和生产订单；请求、承诺、发运和收货日期；分批收货；供应商日历；最小量及包装规格。
Safety stock, target service, review frequency, reorder rules, shelf life, carrying cost, stockout consequence, capacity, and approved overrides.
安全库存、目标服务、检查频率、订货规则、保质期、持有成本、缺货后果、容量和已批准调整。
Reconcile opening inventory to the system of record and timestamp the extract. Correct stockout-censored sales when possible; low sales during no availability are not evidence of low demand. Keep forecast versions so accuracy can be measured against what was genuinely known at each decision date.
应把期初库存与记录系统核对并为数据提取加时间戳。尽可能修正受缺货压制的销量；无库存期间销量低并不证明需求低。保留预测版本，才能用各决策日真实可获得的信息衡量准确性。
Choose inventory forecasting methods by demand pattern根据需求形态选择库存预测方法
The inventory projection itself uses stock-flow arithmetic, but its demand input can come from different methods. Select and compare methods at the decision horizon rather than choosing the most complex model by reputation.
库存推演本身使用库存流量计算，但其需求输入可以来自不同方法。应在实际决策周期内选择并比较方法，而不是因声誉而选择最复杂的模型。
| Method方法 | Useful when适用场景 | Watch for注意事项 |
|---|---|---|
| Naive or seasonal naive朴素或季节朴素法 | Baseline, stable items, or repeated seasonality基线、稳定物料或重复季节性 | Structural change and promotions结构变化和促销 |
| Moving average or exponential smoothing移动平均或指数平滑 | Regular demand with manageable level, trend, or seasonality具有可管理水平、趋势或季节性的规律需求 | Lag after abrupt changes and parameter drift突变后的滞后和参数漂移 |
| Causal or machine-learning model因果或机器学习模型 | Reliable price, promotion, calendar, channel, or external drivers exist存在可靠价格、促销、日历、渠道或外部驱动 | Leakage, unstable features, explanations, and maintenance信息泄漏、不稳定特征、解释和维护 |
| Intermittent-demand or empirical method间歇需求或经验方法 | Many zero periods, lumpy orders, or low-volume parts大量零需求期、块状订单或低量零件 | Normal assumptions and percentage-error metrics may fail正态假设和百分比误差指标可能失效 |
| Judgmental adjustment判断性调整 | Launches, discontinuations, campaigns, known contracts, or disruptions新品、退市、活动、已知合同或中断 | Require reason codes, baseline comparison, owner, and expiry需要原因代码、基线比较、责任人和失效日期 |
Forecast demand separately from confirmed orders when both matter, and define consumption logic to prevent double counting. For example, a customer order may replace part of the baseline forecast rather than add to it. The method must reflect whether lost sales, backorders, or substitutions are recorded.
当预测需求和确认订单都重要时，应分别保存并定义冲销逻辑，避免重复计算。例如，客户订单可能替代部分基线预测，而不是在其上叠加。方法还必须反映是否记录丢失销售、延期交货或替代。
How to build an inventory forecast step by step如何逐步建立库存预测
- Define the decision定义决策Specify users, SKU-location scope, horizon, bucket, refresh cadence, service objective, and whether the output is advisory or feeds an approved planning process.明确用户、SKU-地点范围、预测期、时间桶、刷新频率、服务目标，以及输出是建议还是进入已批准计划流程。
- Create an as-of snapshot创建时点快照Freeze opening stock, allocations, open orders, usable receipts, demand versions, policy parameters, and master data with a common timestamp.用统一时间戳冻结期初库存、分配、未结订单、可用收货、需求版本、政策参数和主数据。
- Generate and reconcile demand生成并核对需求Produce a baseline, add documented business intelligence, consume confirmed orders correctly, and reconcile hierarchy totals before projecting inventory.生成基线，加入有记录的业务信息，正确冲销确认订单，并在推演库存前核对层级总量。
- Project each period逐期推演Roll beginning inventory forward using dated receipts and demand. Flag the first breach of zero, safety stock, capacity, shelf life, or another approved threshold.使用有日期的收货和需求滚动期初库存，标记首次跌破零、安全库存、容量、保质期或其他已批准阈值的时间。
- Run scenarios and publish运行场景并发布Compare demand, receipt, lead-time, and policy scenarios; document overrides, approval, constraints, and the forecast version used for later measurement.比较需求、收货、提前期和政策场景，记录调整、审批、约束以及供后续衡量使用的预测版本。
Simple inventory forecast and net-requirement calculator简单库存预测与净需求计算器
This calculator illustrates an aggregate horizon. It assumes all entered receipts and any new requirement are usable within that horizon; a production forecast must phase quantities by date, lead time, capacity, and approval status.
本计算器用于演示一个汇总预测期，假设所输入收货和新增需求都能在期内使用。生产环境必须按日期、提前期、产能和审批状态对数量分期。
Hypothetical example: 12 periods × 120 units gives 1,440 forecast units. With 600 usable on hand, 500 usable receipts, and a 250-unit target ending stock, net requirement is max(0, 1,440 + 250 − 600 − 500) = 590 units. The negative pre-order ending inventory shows the aggregate shortage, but not its first date. A time-phased schedule is required before action.
假设示例：12 个期间 × 120 件得到 1,440 件预测需求。可用在手库存 600 件、可用收货 500 件、目标期末库存 250 件，则净需求为 max(0, 1,440 + 250 − 600 − 500) = 590 件。新增订货前的负期末库存显示汇总短缺，却不能说明首次短缺日期；行动前必须建立分期计划。
Use inventory forecast scenarios, not one deterministic line使用库存预测场景，而不是单一确定曲线
A single expected-value forecast hides risk. Build a small set of governed scenarios rather than dozens of decorative variants. A baseline uses the approved demand and confirmed supply assumptions. A demand-up scenario tests forecast uncertainty or promotion lift. A supply-delay scenario moves receipts according to plausible lead-time disruption. A policy scenario changes safety stock, service target, order quantity, or review cadence.
单一期望值预测会掩盖风险。应建立少量受治理场景，而不是大量装饰性变体。基线使用已批准需求和确认供应假设；需求上行情景测试预测不确定性或促销提升；供应延迟情景按合理提前期中断移动收货；政策情景调整安全库存、服务目标、订货量或检查频率。
First projected shortage date, duration below safety stock, units short, affected customers or production, and feasible recovery date.
首次预计短缺日、低于安全库存的持续时间、短缺数量、受影响客户或生产以及可行恢复日期。
Peak inventory, capacity breach, days of supply, expiry or obsolescence exposure, markdown, storage, and working-capital implications.
库存峰值、容量超限、供应天数、过期或过时风险、降价、仓储和营运资金影响。
Compare scenarios at the same as-of date, grain, horizon, and policy version. State whether supply is confirmed, planned, or merely assumed. A scenario does not become a commitment until it passes feasibility, economics, approval, and execution controls.
场景必须在相同时点、粒度、预测期和政策版本下比较，并说明供应是确认、计划还是仅假设。只有通过可行性、经济性、审批和执行控制后，场景才能成为承诺。
Common inventory forecasting errors and special cases常见库存预测错误与特殊场景
- Using sales as unconstrained demand: stockouts suppress sales and can create a repeating under-forecast.把销量当作未受约束需求：缺货会压制销量并导致持续低估。
- Counting every open order: late, cancelled, quality-blocked, duplicated, or allocated supply may not be usable.计入所有未结订单：延迟、取消、质检冻结、重复或已分配的供应可能不可用。
- Ignoring timing: equal monthly totals do not prevent a mid-month stockout or capacity breach.忽略时间：月度总量相等并不能防止月中缺货或容量超限。
- Double-counting orders and forecasts: define forecast consumption at item, customer, channel, and time levels.重复计算订单与预测：应在物料、客户、渠道和时间层级定义预测冲销。
- One model for every item: intermittent parts, launches, seasonal items, promotions, end-of-life stock, perishables, and dependent demand need differentiated treatment.所有物料用同一模型：间歇零件、新品、季节品、促销品、退市库存、易腐品和相关需求需要差异化处理。
- Optimizing forecast accuracy alone: a small statistical gain may not improve service, inventory, waste, or working capital.只优化预测准确率：统计误差小幅改善不一定改善服务、库存、浪费或营运资金。
Negative projected inventory can be a useful shortage signal; hiding it at zero removes the magnitude of unmet need. Keep both physical on-hand and unconstrained projected balance when the business needs to distinguish backorders, lost sales, and service impact.
预计库存为负可以作为有用的短缺信号；简单截断为零会丢失未满足需求的幅度。当业务需要区分延期交货、丢失销售和服务影响时，应同时保留实物在手库存和不受约束的预计余额。
Validate inventory forecasts with operational outcomes结合运营结果验证库存预测
Backtest with rolling decision dates and no future leakage. At each historical as-of date, rebuild the opening inventory, demand forecast, known orders, expected receipts, lead-time assumptions, and policy that were actually available. Project forward, then compare with realized inventory and service. Evaluate by horizon and segment rather than only one company-wide average.
应使用滚动决策日且无未来信息泄漏的方式回测。在每个历史时点，重建当时真实可用的期初库存、需求预测、已知订单、预计收货、提前期假设和政策，再向前推演并与实际库存和服务比较。应按预测期与分群评估，而不是只看公司总体平均。
MAPE is unstable when actual demand is zero or very small, so do not make it the only metric for intermittent inventory. Bias reveals systematic over- or under-forecasting; WAPE summarizes absolute error across a group; inventory and service outcomes show whether the forecast was decision-useful. A change in outcomes is not automatically caused by the model—promotions, supply disruptions, assortment, prices, and policy may change simultaneously.
当实际需求为零或很小时，MAPE 不稳定，因此不能把它作为间歇库存的唯一指标。偏差揭示系统性高估或低估，WAPE 汇总一组物料的绝对误差，库存与服务结果则说明预测是否对决策有用。结果变化不一定由模型造成，因为促销、供应中断、商品组合、价格和政策可能同时变化。
Implement a governed inventory forecasting process实施受治理的库存预测流程
Start with a representative pilot across stable, seasonal, intermittent, high-value, and constrained items. Freeze a baseline, define acceptance criteria, and publish a forecast calendar. Assign owners for demand data, inventory accuracy, open-order quality, lead-time definitions, policy parameters, model performance, overrides, approvals, and system changes.
应从覆盖稳定、季节性、间歇、高价值和受约束物料的代表性试点开始。冻结基线、定义验收标准并发布预测日历。为需求数据、库存准确性、未结订单质量、提前期定义、政策参数、模型表现、调整、审批和系统变更指定责任人。
After launch, monitor data freshness, reconciliation breaks, missing forecast snapshots, receipt-date drift, inventory-adjustment frequency, override size and value added, forecast bias, shortage-date accuracy, realized service, excess, expiry, and planner workload. Use thresholds to trigger review; do not automatically overwrite policies merely because one period changes.
上线后，应监控数据新鲜度、核对差异、缺失预测快照、收货日期漂移、库存调整频率、调整幅度与增值、预测偏差、短缺日期准确性、实际服务、过剩、过期和计划人员工作量。使用阈值触发复核，不要因为一个期间变化就自动覆盖政策。
Revisit the design after supplier, lane, assortment, warehouse, production, calendar, service-policy, or unit-of-measure changes. Archive every published version with its as-of timestamp and assumption set so later analysis can distinguish a bad forecast from changed execution.
供应商、线路、商品组合、仓库、生产、日历、服务政策或计量单位变化后，应重新审视设计。归档每个已发布版本及其时点和假设集，才能在后续分析中区分预测失误与执行变化。
Analyze inventory forecasts across connected business data跨关联业务数据分析库存预测
Prepare forecast snapshots, demand history, stockout flags, on-hand balances, allocations, purchase and transfer orders, receipt events, supplier lead times, safety-stock policy, service results, and inventory costs. InfiniSynapse can be used as an analysis layer to explore connected data, compare SKU-location segments, investigate deviations, and review observed outcomes around an existing forecasting process.
请准备预测快照、需求历史、缺货标记、在手余额、分配、采购与调拨订单、收货事件、供应商提前期、安全库存政策、服务结果和库存成本。InfiniSynapse 可作为分析层，用于探索关联数据、比较 SKU-地点分群、调查偏差，并审查现有预测流程周边的实际结果。
Capability boundary: this page does not present InfiniSynapse as a replenishment, purchasing, allocation, warehouse, production-planning, or transport-execution system. It does not automatically create or approve orders or write transactions back to ERP.
能力边界：本页不把 InfiniSynapse 描述为补货、采购、分配、仓库、生产计划或运输执行系统。它不会自动创建或批准订单，也不会向 ERP 回写交易。
Open the InfiniSynapse analytics workspace打开 InfiniSynapse 在线分析工作区Inventory forecasting FAQ库存预测常见问题
What is inventory forecasting?什么是库存预测？
It estimates how stock changes over future periods using expected demand, current inventory, scheduled receipts, lead times, policy buffers, and known events.
它使用预期需求、当前库存、计划收货、提前期、政策缓冲和已知事件，估算未来各期间库存如何变化。
What is the basic inventory forecasting formula?基本库存预测公式是什么？
Projected ending inventory equals beginning inventory plus usable receipts minus demand. A simplified net requirement covers forecast demand plus target ending stock after subtracting available inventory and receipts.
预计期末库存等于期初库存加可用收货减需求。简化净需求是在扣除可用库存和收货后，覆盖预测需求与目标期末库存所需的数量。
How is inventory forecasting different from demand forecasting?库存预测与需求预测有什么区别？
Demand forecasting estimates future demand. Inventory forecasting combines that estimate with stock, supply, timing, and policy inputs to project inventory positions and risk.
需求预测估计未来需求；库存预测把它与库存、供应、时间和政策输入组合，推演库存位置与风险。
How often should an inventory forecast be updated?库存预测应多久更新一次？
Align refresh with decision speed and data freshness, and recalculate after material demand, receipt, supplier, lead-time, assortment, inventory-accuracy, promotion, or policy changes.
刷新频率应匹配决策速度与数据新鲜度，并在需求、收货、供应商、提前期、商品组合、库存准确性、促销或政策发生重大变化后重新计算。
Is InfiniSynapse an inventory replenishment system?InfiniSynapse 是库存补货系统吗？
It is positioned here as an analysis layer for connected forecasting and inventory data, not as the system that creates purchase orders, allocates stock, operates warehouses, schedules transport, or writes ERP transactions.
本页将其定位为分析关联预测与库存数据的分析层，而不是创建采购订单、分配库存、运营仓库、安排运输或回写 ERP 交易的系统。
Official inventory forecasting references and limitations库存预测官方参考与限制
Microsoft explains that its inventory forecast represents the balance between entered supply and demand for an item and helps teams review expected changes to on-hand inventory in its Dynamics 365 inventory forecast documentation. SAP documents projected stock as a period-end demand-and-stock balance calculated from stocks, receipts, and demands in its Calculation of Projected Stock.
Microsoft 在其 Dynamics 365 库存预测文档中说明，库存预测表示物料已输入供应与需求之间的平衡，并帮助团队审查预期在手库存变化。SAP 在预计库存计算文档中将预计库存说明为由库存、收货和需求计算出的期末需求-库存平衡。
Actual planning formulas vary with product configuration, demand consumption, constraints, statuses, calendars, and policy. Verify current system documentation and business definitions. This page is educational and does not replace statistical, financial, accounting, procurement, planning, or inventory-control advice. For monitoring principles, see InfiniSynapse's live guide to a focused data dashboard.
实际计划公式会随产品配置、需求冲销、约束、状态、日历和政策变化，应核验当前系统文档与业务定义。本页仅供教育参考，不替代统计、财务、会计、采购、计划或库存控制建议。关于监控原则，可参阅 InfiniSynapse 已上线的聚焦型数据仪表板指南。