Six-step cleaning of historical cost data and three cost-forecasting paths (price-source classification / four-dimension tagging / index adjustment)
Historical cost data cannot be used directly for forecasting: first apply six-step cleaning (source classification, completeness and deduplication, outlier handling, caliber normalization, four-dimension tagging, average-level adjustment), then choose a forecasting path — the consumption-indicator + current-price method is the most stable; the index-adjustment method uses base date 2017-01-01 with index 1000; use the analogy method when samples are insufficient; mixing market prices/published prices/vendor quotations, different cost types, or different regions is strictly forbidden.
Applicable Codes
- GB/T 51290-2018 建设工程造价指标指数分类与测算标准
- T/ZBD 100.1-2022 建设工程造价数据建设通用标准
- DBJ/T 15-153-2019 广东省建设项目全过程造价管理规范
Topic Positioning
This card provides an actionable method of "historical cost data cleaning -> cost forecasting": the six-step cleaning process, three forecasting paths, and database-building requirements, applicable to cost-indicator measurement and cost forecasting based on completed projects' settlement/budget/quotation data.
Core Conclusions
Six cleaning steps: (1) source classification — market prices (mainly transaction prices), published (information) prices, and vendor quotations are stored strictly by channel, never mixed (T/ZBD 100.1-2022, Clause 7.2.3); (2) completeness check and deduplication; (3) outlier handling — statistical methods (box plot / 3-sigma) plus engineering judgment, distinguishing entry errors (correct/delete) from genuinely special projects (keep and tag); (4) caliber normalization — unify cost scope, measurement rules, and cost type; (5) four-dimension tagging — region (down to county/district), project type, cost type, time (GB/T 51290-2018, Clause 3.0.3); (6) adjustment to the corresponding average level (Clause 5.2.4). Time basis: estimates/approximate estimates/tender control prices use the compilation completion date, contract prices the commencement date, settlement prices the completion date (Clause 3.0.2). Three forecasting paths: (1) consumption indicator + current price — decoupling price from physical consumption, the most stable, so cleaning prioritizes preserving consumption data; (2) index adjustment — base date January 1, 2017 with index 1000; current-period index = current-period indicator / base-period indicator x 1000 (Clause 6.0.2); (3) analogy — fallback when samples are insufficient. At least 10 comparable samples of the same type are recommended; cleaning actions are logged and traceable; results are entered into an information database supporting retrieval, matching, indicator adjustment, and analysis (DBJ/T 15-153-2019, Clause 3.7.4).
Applicable Boundaries
Applies to normalized reuse of historical cost data across periods and projects; not applicable to single-point price queries. Data of different cost types (estimate != approximate estimate != control price != settlement price), different regions, or different periods must never be mixed directly; published prices and vendor quotations deviate from actual markets and serve only as classified references. With too few samples, statistical significance is insufficient and the result should be downgraded to a rough analogy estimate.
Further Reading
Cleaning and database basis: GB/T 51290-2018 Clauses 3.0.2, 3.0.3, 5.2.4, 6.0.2; T/ZBD 100.1-2022 Clause 7.2.3; DBJ/T 15-153-2019 Clauses 3.7.4, 8.3.1~8.3.4. Indicator compilation must follow the principles of completeness/accuracy/applicability, unified cost allocation, and consistent indicator scope (Clause 8.3.4).
Keywords
- 造价数据清洗
- 历史数据
- 成本预测
- 造价指标指数
- 消耗量指标
- market price
- published price information (government guidance prices)
- 厂商报价