料衡 Feedy

数据与方法

说明 Feedy Database v4 的数据来源、字段口径、来源专属指标、批次估算和结果使用边界。

1. Database v4 覆盖范围

当前 Feedy Database v4 共收录 856 条来源记录:美国 NRC 2012 为 122 条,法国 INRA/AFZ 为 127 条,荷兰 CVB 2023 为 385 条,GB/T 39235-2020《猪营养需要量》为 222 条。v4 在保留原有常用饲料原料的基础上,补入了来源资料中此前未进入网站的油脂、矿物质、氨基酸添加剂、纯化合物、微量元素和维生素来源等记录。

2. 来源名称与原料归类分开

每条记录分别保存来源原名、来源编号、family、spec、conceptId 与加工身份。来源名称用于忠实还原原始数据库,family/spec 用于网站检索与归类,两者不互相覆盖。NRC 中文名称以本站使用的 NRC 中文版为准;英文名称独立保存用于身份核对。

3. 源数据、派生值、0 与空白

v4 区分 source_direct(来源直接值)derived(按明确公式换算/派生)。源表明确写 0 时保存为 0;源表空白或未提供时保存为 null,不使用同类原料、其他数据库或模型值去填充 reference 层。样本量 n、SD、原始单位、换算公式和来源 warning 在资料可得时一并保留。

4. 不强行统一不同数据库的营养体系

五类营养数据来源存在不同的指标定义。NRC 与 GB/T 的 ATTD/STTD 磷、INRA 的磷消化率体系、CVB 的 IP/StaDCP/StaDP 均按来源分别保存;氨基酸同样区分总量、AID/SID 系数及可消化含量。脂肪指标中的 EE、AEE、CVB CFAT/CFATh 也不会在底层合并成一个含义模糊的字段。网站提供统一核心视图,但完整来源详情保留数据库原始口径。

5. 网站兼容层与完整来源详情

本候选保留 Release L 已验证的 INRA/CVB 净能模型、采购价值、成本优化和生长参考逻辑;营养速查以 Database v4 四库 856 条兼容层为基线,并追加 297 条 FEDNA 独立参考记录,当前查询层共 1153 条。四库中不具备成本模型所需完整营养组合的纯矿物质、维生素来源等记录仍仅供查询;FEDNA 记录则统一经“本批原料”层后才可进入后续决策,不允许理论参考值直接导入采购价值或成本优化。四库 nested master 与 FEDNA master 均在用户打开“查看完整来源指标”时按需加载,以减少首页初始下载量。

6. 本批原料与系统预测

数据库值代表参考原料。建立“本批原料”后,可录入实际掌握的水分、粗蛋白、灰分、脂肪、纤维等指标。最终取值遵循:手工覆盖 > 本批实测 > 系统预测 > 数据库参考。猪净能是批次评估的核心预测结果;CF、淀粉、NDF、ADF 等模型补全值仅为计算辅助,并不等同于实验室检测结果。

7. SID 氨基酸批次调整

本批 SID Lys、Met、Met+Cys、Thr、Trp、Val、Ile 可按本批粗蛋白与数据库粗蛋白比例作快速初估,并允许逐项手工覆盖。该比例法是 Feedy 的批次快速估算规则,不代表四个来源数据库共享同一官方回归公式。

8. 国外加工副产物与结果解释

对 middlings、shorts、feed meal、feed flour、screenings、hominy feed、expeller、solvent extracted、DDG/DDGS、pulp、solubles 等记录,Feedy 保留原始加工流身份,并补充国内对应关系;不存在严格一一对应商品时,不强行翻译成某个国内商品。

9. FEDNA 猪阶段建议添加量

Feedy V4.10 新增独立的 FEDNA Annex II 实践参考层。它不改变当前 856 条营养来源记录,也不参与 Release L 净能、采购价值或成本优化计算。页面按教槽/早期仔猪(<32 d)、仔猪(33–70 d)、生长猪(>70 d)、妊娠和哺乳阶段展示 FEDNA 建议最大添加比例;F 与 NA 保留 FEDNA 原始定义,脚注和近似身份映射单独提示。该模块属于行业实践参考,不是法规限量或绝对营养上限。

净能、采购价值、配方成本试算和生长参考用于专业分析与决策辅助。实际采购、配方调整和生产应用应结合本批检测、企业标准、配方约束、动物阶段和生产验证。来源资料本身存在异常值时,Feedy 原值保留并以 warning 标识,而不是静默修正。

Database v4 四库基线 856 条 · V4.500 候选查询层 1153 条(含 FEDNA 297 条)· Release L 计算核心保持冻结 · 更新日期:2026-09-06 · 联系邮箱:313370213@qq.com

Data & Methodology

How Feedy Database v4 handles source records, field definitions, source-specific metrics, batch estimation, and interpretation limits.

1. Database v4 coverage

Feedy Database v4 contains 856 source records: 122 from U.S. NRC 2012, 127 from France INRA/AFZ, 385 from Netherlands CVB 2023, and 222 from GB/T 39235-2020 Nutrient Requirements of Swine. In addition to the previously available feed ingredients, v4 adds source records for fats and oils, minerals, amino-acid additives, pure compounds, trace-mineral products and vitamin sources that were previously omitted from the website.

2. Source identity is separate from classification

Each record keeps its source name, source code, family, specification, concept ID and processing identity. Source names reproduce the original database; family/spec are used for Feedy search and classification and do not overwrite source identity. NRC Chinese names follow the Chinese edition used for this database build, while English names are retained independently for identity checking.

3. Source values, derived values, zero and null

v4 distinguishes source_direct values from derived values calculated by an explicit formula. A source-reported zero remains zero; an unreported/blank source cell remains null. Feedy does not fill the reference layer from similar ingredients, another database or a model. Where available, sample n, SD, original units, conversion formulae and source warnings are retained.

4. Source-specific nutrition systems are preserved

The five source families now exposed in the lookup layer use different definitions. NRC/GB/T ATTD-STTD phosphorus, INRA phosphorus digestibility and CVB IP/StaDCP/StaDP are stored separately. Amino acids distinguish total content, AID/SID coefficients and digestible content. EE, AEE and CVB CFAT/CFATh are likewise retained as distinct source fields. Feedy provides a common comparison view without erasing the original system.

5. Compatibility layer and full source detail

This candidate keeps the validated Release L INRA/CVB net-energy models, value assessment, cost optimisation and growth-reference logic. The nutrition lookup uses the 856-record four-source Database v4 compatibility layer as its baseline and adds 297 independent FEDNA references, for 1,153 lookup records in total. Four-source pure minerals, vitamin sources and other records without the complete nutrient combination required by the cost model remain query-only; FEDNA references must pass through the Batch Ingredient layer before downstream decisions and cannot directly enter procurement-value or cost-optimisation workflows. The nested four-source and FEDNA masters are loaded on demand when “Full source metrics” is opened.

6. Batch ingredients and prediction

Database values represent reference ingredients. Batch records can contain measured moisture, crude protein, ash, fat, fibre and other values. Final-value priority is manual override > batch measurement > model estimate > database reference. Pig net energy is the main batch prediction output; model-completed CF, starch, NDF and ADF are calculation aids, not laboratory measurements.

7. Batch SID amino-acid scaling

SID Lys, Met, Met+Cys, Thr, Trp, Val and Ile may be provisionally scaled by the batch-CP/reference-CP ratio and manually overridden. This is a Feedy batch-screening rule, not a claim that all source databases use one universal official regression.

8. Processing by-products and interpretation

For middlings, shorts, feed meal, feed flour, screenings, hominy feed, expeller, solvent extracted, DDG/DDGS, pulp and solubles, Feedy retains the original processing-stream identity and provides domestic-equivalence notes where useful. It does not force a one-to-one Chinese commodity name when no strict equivalent exists.

9. FEDNA stage-wise practical inclusion guidance

Feedy V4.10 adds FEDNA Annex II as a separate practical-reference layer. It does not alter the 856 nutrition source records and does not feed into the Release L net-energy, procurement-value or cost-optimisation models. The interface keeps the FEDNA stages for Prestarter (<32 d), Starter (33–70 d), Grower (>70 d), Gestation and Lactation, while preserving the source meanings of F and NA, source footnotes, and explicit warnings for approximate identity mappings. These are practical reference limits, not statutory limits or absolute nutritional constraints.

Net-energy, procurement value, cost optimisation and growth-reference outputs are decision-support tools. Procurement, formulation changes and production use should also consider batch testing, company standards, formulation constraints, animal stage and production validation. Apparent anomalies in source publications are retained with a warning rather than silently altered.

Database v4 four-source baseline: 856 records · V4.500 candidate lookup: 1,153 records including 297 FEDNA references · Release L calculation core remains frozen · Updated: 6 Sep 2026 · Contact: 313370213@qq.com

FEDNA 营养参考(V4.500 候选)

四库 Database v4 基线仍为 856 条。本候选在其上新增 FEDNA 2021 营养参考 297 条(157 条标准矩阵 + 140 条专项记录),候选营养查询层合计 1153 条。FEDNA 记录可用于营养查询和“本批原料”参考,但不允许直接导入采购价值或成本优化;决策入口必须先经过本批原料层。Annex II 使用量参考继续作为独立约束层。

FEDNA Nutrition References (V4.500 Candidate)

The four-source Database v4 baseline remains 856 records. This candidate adds 297 FEDNA 2021 nutrition references (157 standard matrices + 140 specialty records), bringing the candidate nutrition lookup layer to 1153 records. FEDNA records may be used for nutrition lookup and as Batch Ingredient references, but they cannot be imported directly into procurement-value or cost-optimisation workflows. Annex II inclusion guidance remains a separate reference layer.

FEDNA 来源复核补记(V4.210)

对 FEDNA 原书 pp.536、553–554 做第二轮源表复核后,补入 p554 最后一列 L-Isoleucine(UE 3c383,90% DM);同时恢复此前遗漏的 K 行、MHA 88.05% 原值、赖氨酸硫酸盐 P/硫酸根信息,并对 TBCC 未标注“15”、硫酸锌水合物表头/化学式不一致、L-Histidine HCl 鱼类授权脚注保留 warning。上述均属于来源忠实度修正,不改变 Release L 计算模型。

来源详情增强(V4.210)

FEDNA 专项矿物质、微量元素和工业氨基酸继续补充 CAS、EU/目录编号、来源表名、PDF 页码与源表脚注。对于 MgO,主表直接值保留为 >50%,同时另存脚注“50–55%”范围,不再用脚注范围覆盖主表原值。

FEDNA 标准矩阵来源完整性(V4.210)

对 157 条 FEDNA 标准营养矩阵重新按 PDF 源表解析宏量矿物质、微量元素及磷体系,共核对 2,826 个来源单元,数值不一致为 0。同时发现并修复 FEDNA-037、FEDNA-038 因紧凑表头 Pphytic / Pavail. 未被旧解析器识别而遗漏的 22 个查询字段。完整来源详情现在可显示 Ca、P、植酸磷、可利用磷、禽/猪消化磷、Na、Cl、Mg、K、S,以及 Cu、Fe、Mn、Zn 等来源值。

跨库差异处理原则(V4.210)

对当前筛出的 107 个 FEDNA 与其他数据库高差异项逐项回到 FEDNA 来源表核对;本轮均确认与来源记录一致,因此标记为 source_verified_difference。跨库差异只作为 QA 与解释提示,不自动修改任何数据库值。

计量基础与本批原料入口(V4.210)

营养结果概览新增“计量基础”列;当所选记录混合 as-fed、来源专项表口径,或包含 p554 干物质基础氨基酸时,界面会显式提示口径差异。FEDNA 的 batch_only 记录不再只显示灰色禁用按钮,而是提供“建立本批原料”入口;其理论值仍不能直接导入采购价值或成本优化。

FEDNA Annex II 身份绑定(V4.210)

V4.10 时 Annex II 共 110 行,其中 83 行能够绑定到四库 856 条 canonicalIngredient。随着 FEDNA 营养参考层补入并完成新增原料身份审计,本候选可安全启用另外 20 行绑定,使当前可绑定数量提升到 103 / 110。这一步只更新“约束行 ↔ canonicalIngredient”的身份连接,不改变 FEDNA Annex II 的任何阶段数值或脚注;其余 7 行仍保持未绑定。

FEDNA 专项源表语义复核(V4.220)

本轮对 101 条 FEDNA 专项记录按原书表头和脚注做结构/语义复核。重点修正 p515 钙源消化率列语义:猪栏依次为 ATTD、STTD、TTTD、AID、TID,肉鸡栏为 ATTD、TTTD;此前候选把猪 AID/TID 误标为禽 AID/TID,本轮只修字段语义并补回来源已有的禽 ATTD/TTTD,不改任何源数值。

油脂质量与鱼油双表来源(V4.220)

补充 FEDNA 油脂专项的 Titer、皂化值、MIU、NEM、矿物酸度、灰分和 Ca 等来源质量指标(仅来源有值时展示)。鱼油 p490 的商品/地区列与 p492 的鱼种参考表分开保存:两张表数值并不完全相同,因此不互相覆盖。

工业氨基酸来源口径(V4.220)

p553 的来源表脚注为“Values expressed as is”;p554 则明确写明“除能量值外均为干物质基础”。本候选将 p554 的该口径应用到全部 8 条记录,并补全来源给出的物理形态、水分、粗蛋白(N×6.25)、灰分、总能、禽 AMEn 与兔 DE 等信息。粗蛋白仍只作为来源字段,不作为合成氨基酸的主判断指标。

FEDNA 磷源表完整性补充(V4.230)

除 p526 的“P 消化率/相对利用率汇总表”外,FEDNA 还在 p527–528 给出磷酸盐原料组成表。本轮将可明确对应的 7 个既有磷源记录补入化学式、目录号、CAS(有值时)、水分/灰分、Ca/P/Na/K/Cl/Mg/S 及可明确读取的其他元素,并新增脱胶骨粉、水合磷酸镁、水合钠钙镁复合磷酸盐、磷酸二氢铵 4 个组成型来源记录。p524–525 的文献综述 ATTD/STTD/相对利用率与 p526 商业来源汇总分层保存,互不覆盖。

注意:p527 英译表本身存在标签/排版异常(11.3.x 编号出现在“CAS nº”行、CAS 格式值出现在“Catalog nº”行,表尾又出现第二个“Moisture”行)。Feedy 保存原始标签与数值并显示 warning,不把可疑尾部三行静默改写成 Fluor/Fe/Cu。

FEDNA 添加剂/纯品专项扩展(V4.240)

继续按 FEDNA 原书专项表完整性,而不是按早期目标条数截断:新增 p541 甜菜碱 3 条、p545 淀粉/糖 5 条、p546 有机酸/醇 11 条、p547 有机酸盐 9 条、p548 氯化胆碱 4 条、p556 非蛋白氮 3 条。所有记录仍是 FEDNA 独立参考层,禁止直接进入采购价值/配方成本模型。

物种边界单独保留:例如 FEDNA 对尿素明确限定“具有功能性瘤胃的动物”,因此在猪料网站中只提供来源查看,不把其总能、N 或蛋白当量用于猪营养决策。

FEDNA 方法与使用说明层(V4.260)

新增独立 guidance 层保存甜菜碱、胆碱、有机酸、非蛋白氮、Se/Co/Mo 等来源叙述,并保存 Annex I 猪能值预测公式。该层 reference-only / non-executable:用于解释 FEDNA 的来源口径,不调用、不覆盖 Release L 净能、采购价值和成本优化计算。

Annex I 的 DEg、ME/DE、NE 公式按 FEDNA p561 保存为方法参考;其中原料类别对应的消化率假设仍需按 FEDNA 原文判断,系统不把它们自动套到“本批原料”。

本批原料数据库漂移与备份安全(V4.300)

本批原料保存时同时记录数据库参考快照、创建版本和最后更新版本。以后若同一 referenceRecordId 的数据库参考发生变化,Feedy 不再在页面刷新时静默重算历史批次,而要求用户明确选择“使用保存快照”或“按当前数据库临时重算”;该选择只影响当前决策读取,不回写历史本批记录。若原 referenceRecordId 已消失,历史批次继续保留,不按同名原料或第一条候选自动映射;需要编辑时必须显式选择新的数据库参考,并建立新的本批记录。

本批备份升级到 v3,分别保存 UI 版本、UI 数据版本、Release L 核心模型版本和核心数据版本。导入支持严格恢复、安全合并和智能合并;智能合并只在备份 updatedAt 明确晚于本机版本时替换,时间相同或无法判断时保留本机。严格恢复和智能合并前自动保存本机备份。

Batch database drift and backup safety (V4.300)

Each Batch Ingredient stores its reference snapshot plus creation/update version provenance. If the same referenceRecordId later changes, Feedy no longer silently recomputes the historical batch on refresh. The user must explicitly choose the saved snapshot or a transient recomputation from the current database; this decision is read-only and does not rewrite the stored historical batch. If the original record ID disappears, the historical batch is retained and is never remapped by same name or first candidate. Editing requires an explicit new reference and creates a new batch record.

Batch backups use schema v3 with separate UI version, UI-data version, Release L core-model version and core-data version. Import supports strict restore, safe merge and conservative smart merge; smart merge replaces a local conflict only when the backup updatedAt is explicitly later.

版本维度、交互安全与部署保护(V4.300)

UI 版本、UI 数据版本、Release L 核心模型版本、核心数据版本、生产基线和部署状态由同一 release metadata 提供给 About、备份和诊断,避免历史候选数字冒充当前状态。确认弹窗、统一原料选择器、兼容选择器和本批原料弹窗增加 Tab / Shift+Tab 焦点约束与关闭后焦点返回。

候选包提供 preflight、deploy 和 rollback 脚本。部署前先检查当前资源、FEDNA/四库来源 master、JS 语法和 8 个 Release L 固定 SHA256;写入线上前建立压缩备份和 rollback pointer,写入后任一关键步骤失败自动回滚,手动回滚前也会先备份当前线上目录。该 V4.300 阶段当时仍为 NOT DEPLOYED;当前 V4.500 为生产发布构建。

Version dimensions, interaction safety and deployment protection (V4.300)

About, backup and release diagnostics consume one authoritative metadata object that separates UI/UI-data versions from the frozen Release L core-model/core-data versions, production baseline and deployment status. Confirmation dialogs, ingredient pickers and Batch Ingredient modals add Tab/Shift+Tab focus containment and focus return.

The candidate package includes preflight, deployment and rollback scripts with source-master checks, active-JS syntax checks and eight fixed Release L hashes. Deployment creates a compressed pre-write backup and rollback pointer; a post-write failure triggers automatic rollback, while manual rollback first backs up the current site. That V4.300 stage remained NOT DEPLOYED at the time; current V4.500 is the production release build.

成本试算条件确认

配方成本试算中的目标原料、价格、当前用量或目标用量发生变化后,试算状态自动转为“待确认”。只有用户显式点击“确认试算条件并计算”后,目标条件才进入成本优化计算;未确认的临时目标不会参与模型求解。

本批容量、成本目标与导入安全(V4.400)

本批原料本机上限为 50 条。达到上限后,Feedy 阻止新增与会超限的备份合并,不自动删除最早记录;导入备份若存在重复 batchId 也会阻断,避免系统替用户猜测保留哪一条。编辑本批存在未保存改动时,关闭/按 Esc 必须显式确认放弃。

成本试算继续采用显式目标确认,并新增完整就绪检查:目标价格必须为正数;“提高已有原料”必须有正的当前用量且目标用量严格高于当前用量;“新增原料”必须填写正的目标用量。可调配方原料缺少有效价格时优化器直接阻断,不把空白价格默认为 0。遇到同一原料概念在当前配方存在多个候选时,不自动取第一项,要求用户先明确目标。

Batch capacity, cost-target and import safety (V4.400)

Local Batch Ingredients are capped at 50. At capacity, Feedy blocks new records and over-capacity backup merges instead of deleting the oldest record. Backup imports with duplicate batchId values are blocked rather than guessed. Closing a dirty Batch editor requires explicit discard confirmation.

Cost trials keep explicit target confirmation and add a readiness gate: target price must be positive; increase mode requires a positive current inclusion and a target strictly above it; new mode requires a positive target inclusion. Missing adjustable-ingredient prices block optimisation rather than defaulting blanks to zero. Multiple same-concept formula matches require an explicit user choice rather than a first-item fallback.

V4.500 展示与备份安全说明

V4.500 不改动 Release L 数值模型。FEDNA 功能性添加剂与非蛋白氮来源按用途分型显示;非蛋白氮中的 PDIA/PDIE/PDIN 明确保留为反刍动物来源指标,不转换为猪用值。本批原料管理器可查看最近 3 份破坏性恢复前自动备份,并通过同一恢复预览流程再次恢复。市场价格显示最近更新时间;跨日时仅提示可能过期,不自动阻断历史情景试算。

V4.500 Display & Backup Safety

V4.500 does not alter Release L numerical models. FEDNA functional additives and non-protein-nitrogen sources use type-specific presentation; PDIA/PDIE/PDIN remain explicitly ruminant-source metrics and are not converted into pig values. The Batch Ingredient manager exposes the three most recent automatic pre-restore backups through the same import-preview workflow. Market prices display their last update timestamp; cross-day data are flagged as potentially stale without blocking intentional historical scenario testing.

Search semantics

Feedy V4.4 uses a canonical lookup hierarchy: level-1 category → level-2 category → canonical ingredient → specification → source database record. Direct search resolves a canonical ingredient first rather than treating every record containing the same substring as an equal hit. Nutritional-grade variants remain specifications under the same ingredient where appropriate, while materially different products or processing identities such as fermentation, enzymatic hydrolysis, rumen protection and pressing remain separate canonical ingredients. Original source names and classifications are retained in the record provenance.