Back to Details

ai-ml-timeseries

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

Time Series Forecasting — Modern Patterns & Production Best Practices

时间序列预测 — 现代模式与生产最佳实践

Modern Best Practices (January 2026):
  • Treat time as a first-class axis: temporal splits, rolling backtests, and point-in-time correctness.
  • Default to strong baselines (naive/seasonal naive) before complex models.
  • Prevent leakage: feature windows and aggregations must use only information available at prediction time.
  • Evaluate by horizon and segment; a single aggregate metric hides failures.
  • Prefer probabilistic forecasts when decisions are risk-sensitive (quantiles/intervals); evaluate calibration (coverage) and use pinball/CRPS.
  • For many related series, consider global + hierarchical approaches (shared models + reconciliation); validate across levels and key segments.
  • Treat time zones/DST as first-class; validate timestamp alignment before feature generation.
  • Define retraining cadence and degraded modes (fallback model, last-known-good forecast).
This skill provides operational, copy-paste-ready workflows for forecasting with recent advances: TS-specific EDA, temporal validation, lag/rolling features, model selection, multi-step forecasting, backtesting, generative AI (Chronos, TimesFM), and production deployment with drift monitoring.
It focuses on hands-on forecasting execution, not theory.
现代最佳实践（2026年1月）:
  • 将时间作为核心轴：时间拆分、滚动回测以及时点正确性。
  • 在使用复杂模型前，优先采用强基准模型（朴素/季节性朴素模型）。
  • 防止数据泄露：特征窗口与聚合操作只能使用预测时点可获取的信息。
  • 按预测周期和细分维度评估模型；单一聚合指标会掩盖问题。
  • 当决策对风险敏感时，优先选择概率性预测（分位数/区间）；评估校准度（覆盖率）并使用pinball损失/CRPS指标。
  • 针对多个相关序列，考虑全局+分层方法（共享模型+结果调和）；在不同层级和关键细分维度上验证。
  • 将时区/夏令时作为核心因素；生成特征前验证时间戳对齐情况。
  • 定义重训练周期与降级模式（备用模型、最近的可靠预测值）。
本技能提供可直接复用的操作工作流，涵盖近期技术进展的预测方法：时间序列专属EDA、时间验证、滞后/滚动特征、模型选择、多步预测、回测、生成式AI（Chronos、TimesFM）以及带漂移监控的生产部署。
本技能聚焦实操性预测执行，而非理论知识。

When to Use This Skill

何时使用本技能

Claude should invoke this skill when the user asks for hands-on time series forecasting, e.g.:
  • "Build a time series model for X."
  • "Create lag features / rolling windows."
  • "Help design a forecasting backtest."
  • "Pick the right forecasting model for my data."
  • "Fix leakage in forecasting."
  • "Evaluate multi-horizon forecasts."
  • "Use LLMs or generative models for TS."
  • "Set up monitoring for a forecast system."
  • "Implement LightGBM for time series."
  • "Use transformer models (TimesFM, Chronos) for forecasting."
  • "Apply temporal classification/survival modelling for event prediction."
If the user is asking about general ML modelling, deployment, or infrastructure, prefer:
  • ai-ml-data-science - General data science workflows, EDA, feature engineering, evaluation
  • ai-mlops - Model deployment, monitoring, drift detection, retraining automation
If the user is asking about LLM/RAG/search, prefer:
  • ai-llm - LLM fine-tuning, prompting, evaluation
  • ai-rag - RAG pipeline design and optimization
当用户询问实操性时间序列预测相关问题时，Claude应调用本技能，例如：
  • "为X构建时间序列模型。"
  • "创建滞后特征/滚动窗口。"
  • "帮助设计预测回测方案。"
  • "为我的数据选择合适的预测模型。"
  • "修复预测中的数据泄露问题。"
  • "评估多周期预测结果。"
  • "使用LLM或生成式模型处理时间序列。"
  • "为预测系统设置监控。"
  • "使用LightGBM进行时间序列预测。"
  • "使用Transformer模型（TimesFM、Chronos）进行预测。"
  • "应用时间分类/生存模型进行事件预测。"
如果用户询问通用ML建模、部署或基础设施相关问题，优先使用：
  • ai-ml-data-science - 通用数据科学工作流、EDA、特征工程、评估
  • ai-mlops - 模型部署、监控、漂移检测、重训练自动化
如果用户询问LLM/RAG/搜索相关问题，优先使用：
  • ai-llm - LLM微调、提示工程、评估
  • ai-rag - RAG管道设计与优化

Quick Reference

快速参考

TaskTool/FrameworkCommandWhen to Use
TS EDA & DecompositionPandas, statsmodels
seasonal_decompose()
, 
df.plot()
Identifying trend, seasonality, outliers
Lag/Rolling FeaturesPandas, NumPy
df.shift()
, 
df.rolling()
Creating temporal features for ML models
Model Training (Tree-based)LightGBM, XGBoost
lgb.train()
, 
xgb.train()
Tabular TS with seasonality, covariates
Deep Learning (Sequence models)Transformers, RNNs
model.forecast()
Long-term dependencies, complex patterns
Event forecastingBinary/time-to-event modelsTemporal labeling + rolling validationSparse events and alerts
BacktestingCustom rolling windows
for window in windows: train(), test()
Temporal validation without leakage
Metrics Evaluationscikit-learn, custom
mean_absolute_error()
, MAPE, MASE		Multi-horizon forecast accuracy
Production DeploymentMLflow, AirflowScheduled pipelinesAutomated retraining, drift monitoring
任务工具/框架命令使用场景
时间序列EDA与分解Pandas, statsmodels
seasonal_decompose()
, 
df.plot()
识别趋势、季节性、异常值
滞后/滚动特征Pandas, NumPy
df.shift()
, 
df.rolling()
为ML模型创建时间特征
模型训练（基于树）LightGBM, XGBoost
lgb.train()
, 
xgb.train()
带季节性、协变量的表格型时间序列
深度学习（序列模型）Transformers, RNNs
model.forecast()
长期依赖、复杂模式
事件预测二元/时间到事件模型时间标记+滚动验证稀疏事件与告警
回测自定义滚动窗口
for window in windows: train(), test()
无数据泄露的时间验证
指标评估scikit-learn, 自定义
mean_absolute_error()
, MAPE, MASE		多周期预测准确率
生产部署MLflow, Airflow调度管道自动化重训练、漂移监控

Decision Tree: Choosing Time Series Approach

决策树：选择时间序列方法

text
User needs time series forecasting for: [Data Type]
    ├─ Strong Seasonality?
    │   ├─ Simple patterns? → LightGBM with seasonal features
    │   ├─ Complex patterns? → LightGBM + Prophet comparison
    │   └─ Multiple seasonalities? → Prophet or TBATS
    │
    ├─ Long-term Dependencies (>50 steps)?
    │   ├─ Transformers (TimesFM, Chronos) → Best for complex patterns
    │   └─ RNNs/LSTMs → Good for sequential dependencies
    │
    ├─ Event Forecasting (binary outcomes)?
    │   └─ Temporal classification / survival modelling → validate with time-based splits
    │
    ├─ Intermittent/Sparse Data (many zeros)?
    │   ├─ Croston/SBA → Classical intermittent methods
    │   └─ LightGBM with zero-inflation features → Modern approach
    │
    ├─ Multiple Covariates?
    │   ├─ LightGBM → Best with many features
    │   └─ TFT/DeepAR → If deep learning needed
    │
    └─ Explainability Required (healthcare, finance)?
        ├─ LightGBM → SHAP values, feature importance
        └─ Linear models → Most interpretable
text
User needs time series forecasting for: [Data Type]
    ├─ Strong Seasonality?
    │   ├─ Simple patterns? → LightGBM with seasonal features
    │   ├─ Complex patterns? → LightGBM + Prophet comparison
    │   └─ Multiple seasonalities? → Prophet or TBATS
    │
    ├─ Long-term Dependencies (>50 steps)?
    │   ├─ Transformers (TimesFM, Chronos) → Best for complex patterns
    │   └─ RNNs/LSTMs → Good for sequential dependencies
    │
    ├─ Event Forecasting (binary outcomes)?
    │   └─ Temporal classification / survival modelling → validate with time-based splits
    │
    ├─ Intermittent/Sparse Data (many zeros)?
    │   ├─ Croston/SBA → Classical intermittent methods
    │   └─ LightGBM with zero-inflation features → Modern approach
    │
    ├─ Multiple Covariates?
    │   ├─ LightGBM → Best with many features
    │   └─ TFT/DeepAR → If deep learning needed
    │
    └─ Explainability Required (healthcare, finance)?
        ├─ LightGBM → SHAP values, feature importance
        └─ Linear models → Most interpretable

Core Concepts (Vendor-Agnostic)

核心概念（与厂商无关）

  • Time axis: splits, features, and labels must respect time ordering and availability.
  • Non-stationarity: seasonality, trend, and regime shifts are normal; monitor and retrain intentionally.
  • Evaluation: rolling/expanding backtests; report horizon-wise and segment-wise performance.
  • Operationalization: define retraining cadence, fallback models, and data freshness contracts.
  • Data governance: treat time series as potentially sensitive; enforce access control, retention, and PII scrubbing in logs.
  • 时间轴：拆分、特征与标签必须遵循时间顺序与可用性。
  • 非平稳性：季节性、趋势与机制转变是正常现象；需有意监控并重训练。
  • 评估：滚动/扩展回测；按周期和细分维度报告性能。
  • 落地实施：定义重训练周期、备用模型与数据新鲜度约定。
  • 数据治理：将时间序列视为潜在敏感数据；在日志中强制执行访问控制、保留策略与PII清理。

Implementation Practices (Tooling Examples)

实施实践（工具示例）

  • Build features with explicit time windows; store cutoff timestamps with each training run.
  • Backtest with a standardized harness (rolling/expanding windows, horizon-wise metrics).
  • Log production forecasts with metadata (model version, horizon, data cut) to enable debugging.
  • Implement fallbacks (baseline model, last-known-good, “insufficient data” handling) for outages and anomalies.
  • 使用明确的时间窗口构建特征；为每次训练运行存储截止时间戳。
  • 使用标准化框架进行回测（滚动/扩展窗口、按周期的指标）。
  • 记录生产预测的元数据（模型版本、周期、数据截止时间）以支持调试。
  • 为故障与异常情况实现回退方案（基准模型、最近的可靠值、“数据不足”处理）。

Do / Avoid

注意事项

Do
  • Do start with naive/seasonal naive baselines and compare against learned models (Forecasting: Principles and Practice: https://otexts.com/fpp3/).
  • Do backtest with rolling windows and preserve point-in-time correctness.
  • Do monitor for data pipeline changes (missing timestamps, level shifts, calendar changes).
  • Do align metrics/loss to the decision: asymmetric costs, service levels, and probabilistic targets (quantiles/intervals) when needed.
Avoid
  • Avoid random splits for forecasting problems.
  • Avoid features that use future information (future aggregates, leakage via target encoding).
  • Avoid optimizing only aggregate metrics; always inspect horizon-wise errors and worst segments.
  • Avoid MAPE when the target can be 0 or near-0; prefer MASE/WAPE/sMAPE and horizon-wise reporting.
建议
  • 建议从朴素/季节性朴素基准模型开始，与学习模型进行对比（参考《预测：原理与实践》：https://otexts.com/fpp3/）。
  • 建议使用滚动窗口进行回测，保持时点正确性。
  • 建议监控数据管道变化（缺失时间戳、水平偏移、日历变更）。
  • 建议根据决策调整指标/损失：非对称成本、服务水平，以及必要时的概率目标（分位数/区间）。
避免
  • 避免在预测问题中使用随机拆分。
  • 避免使用包含未来信息的特征（未来聚合、通过目标编码导致的数据泄露）。
  • 避免仅优化聚合指标；务必检查按周期的错误与最差细分维度。
  • 当目标值可能为0或接近0时，避免使用MAPE；优先选择MASE/WAPE/sMAPE并按周期报告。

Navigation: Core Patterns

导航：核心模式

Time Series EDA & Data Preparation

时间序列EDA与数据准备

  • TS EDA Best Practices
    • Frequency detection, missing timestamps, decomposition
    • Outlier detection, level shifts, seasonality analysis
    • Granularity selection and stability checks
  • 时间序列EDA最佳实践
    • 频率检测、缺失时间戳、分解
    • 异常值检测、水平偏移、季节性分析
    • 粒度选择与稳定性检查

Feature Engineering

特征工程

  • Lag & Rolling Patterns
    • Lag features (lag_1, lag_7, lag_28 for daily data)
    • Rolling windows (mean, std, min, max, EWM)
    • Avoiding leakage, seasonal lags, datetime features
  • 滞后与滚动模式
    • 滞后特征（每日数据的lag_1、lag_7、lag_28）
    • 滚动窗口（均值、标准差、最小值、最大值、EWM）
    • 避免数据泄露、季节性滞后、日期时间特征

Model Selection

模型选择

  • Model Selection Guide
    • Decision rules: Strong seasonality → LightGBM, Long-term → Transformers
    • Benchmark comparison: LightGBM vs Prophet vs Transformers vs RNNs
    • Explainability considerations for mission-critical domains
  • LightGBM TS Patterns (feature-based forecasting best practices)
    • Why LightGBM excels: performance + efficiency + explainability
    • Feature engineering for tree-based models
    • Hyperparameter tuning for time series
  • 模型选择指南
    • 决策规则：强季节性→LightGBM，长期依赖→Transformers
    • 基准对比：LightGBM vs Prophet vs Transformers vs RNNs
    • 关键领域的可解释性考量
  • LightGBM时间序列模式 （基于特征的预测最佳实践）
    • LightGBM的优势：性能+效率+可解释性
    • 基于树模型的特征工程
    • 时间序列的超参数调优

Forecasting Strategies

预测策略

  • Multi-Step Forecasting Patterns
    • Direct strategy (separate models per horizon)
    • Recursive strategy (feed predictions back)
    • Seq2Seq strategy (Transformers, RNNs for long horizons)
  • Intermittent Demand Patterns
    • Croston, SBA, ADIDA for sparse data
    • LightGBM with zero-inflation features (modern approach)
    • Two-stage hurdle models, hierarchical Bayesian
  • 多步预测模式
    • 直接策略（每个周期使用独立模型）
    • 递归策略（将预测结果反馈）
    • Seq2Seq策略（Transformers、RNNs用于长周期）
  • 间歇性需求模式
    • Croston、SBA、ADIDA用于稀疏数据
    • 带零膨胀特征的LightGBM（现代方法）
    • 两阶段 hurdle模型、分层贝叶斯模型

Validation & Evaluation

验证与评估

  • Backtesting Patterns
    • Rolling window backtest, expanding window
    • Temporal train/validation split (no IID splits!)
    • Horizon-wise metrics, segment-level evaluation
  • 回测模式
    • 滚动窗口回测、扩展窗口
    • 时间训练/验证拆分（禁止IID拆分！）
    • 按周期的指标、细分维度评估

Generative & Advanced Models

生成式与高级模型

  • TS-LLM Patterns
    • Chronos, TimesFM, Lag-Llama (Transformer models)
    • Event forecasting patterns (temporal classification, survival modelling)
    • Tokenization, discretization, trajectory sampling
  • 时间序列LLM模式
    • Chronos、TimesFM、Lag-Llama（Transformer模型）
    • 事件预测模式（时间分类、生存模型）
    • 分词、离散化、轨迹采样

Production Deployment

生产部署

  • Production Deployment Patterns
    • Feature pipelines (same code for train/serve)
    • Retraining strategies (time-based, drift-triggered)
    • Monitoring (error drift, feature drift, volume drift)
    • Fallback strategies, streaming ingestion, data governance
  • 生产部署模式
    • 特征管道（训练/服务使用相同代码）
    • 重训练策略（基于时间、漂移触发）
    • 监控（错误漂移、特征漂移、量漂移）
    • 回退策略、流摄入、数据治理

Navigation: Templates (Copy-Paste Ready)

导航：模板（可直接复用）

Data Preparation

数据准备

  • TS EDA Template - Reproducible structure for time series analysis
  • Resample & Fill Template - Handle missing timestamps and resampling
  • 时间序列EDA模板 - 可复现的时间序列分析结构
  • 重采样与填充模板 - 处理缺失时间戳与重采样

Feature Templates

特征模板

  • Lag & Rolling Features - Create temporal features for ML models
  • Calendar Features - Business calendars, holidays, events
  • 滞后与滚动特征 - 为ML模型创建时间特征
  • 日历特征 - 业务日历、节假日、事件

Model Templates

模型模板

  • Forecast Model Template - End-to-end forecasting pipeline (LightGBM, transformers, RNNs)
  • Multi-Step Strategy - Direct, recursive, and seq2seq approaches
  • 预测模型模板 - 端到端预测管道（LightGBM、Transformers、RNNs）
  • 多步策略 - 直接、递归与seq2seq方法

Evaluation Templates

评估模板

  • Backtest Template - Rolling window validation setup
  • TS Metrics Template - MAPE, MAE, RMSE, MASE, pinball loss
  • 回测模板 - 滚动窗口验证设置
  • 时间序列指标模板 - MAPE、MAE、RMSE、MASE、pinball损失

Advanced Templates

高级模板

  • TS-LLM Template - Time series foundation model patterns and experimental approaches
  • 时间序列LLM模板 - 时间序列基础模型模式与实验方法

Related Skills

相关技能

For adjacent topics, reference these skills:
  • ai-ml-data-science - EDA workflows, feature engineering patterns, model evaluation, SQLMesh transformations
  • ai-mlops - Production deployment, monitoring, retraining pipelines
  • ai-llm - Fine-tuning approaches applicable to time series LLMs (Chronos, TimesFM)
  • ai-prompt-engineering - Prompt design patterns for time series LLMs
  • data-sql-optimization - SQL optimization for time series data storage and retrieval
对于相邻主题，参考以下技能：
  • ai-ml-data-science - EDA工作流、特征工程模式、模型评估、SQLMesh转换
  • ai-mlops - 生产部署、监控、重训练管道
  • ai-llm - 适用于时间序列LLM（Chronos、TimesFM）的微调方法
  • ai-prompt-engineering - 时间序列LLM的提示设计模式
  • data-sql-optimization - 时间序列数据存储与检索的SQL优化

External Resources

外部资源

See data/sources.json for curated web resources including:
  • Classical methods (statsmodels, Prophet, ARIMA)
  • Deep learning frameworks (PyTorch Forecasting, GluonTS, Darts, NeuralProphet)
  • Transformer models (TimesFM, Chronos, Lag-Llama, Informer, Autoformer)
  • Anomaly detection tools (PyOD, STUMPY, Isolation Forest)
  • Feature engineering libraries (tsfresh, TSFuse, Featuretools)
  • Production deployment (Kats, MLflow, sktime)
  • Benchmarks and datasets (M5 Competition, Monash Time Series, UCI)
查看data/sources.json获取精选网络资源，包括：
  • 经典方法（statsmodels、Prophet、ARIMA）
  • 深度学习框架（PyTorch Forecasting、GluonTS、Darts、NeuralProphet）
  • Transformer模型（TimesFM、Chronos、Lag-Llama、Informer、Autoformer）
  • 异常检测工具（PyOD、STUMPY、孤立森林）
  • 特征工程库（tsfresh、TSFuse、Featuretools）
  • 生产部署（Kats、MLflow、sktime）
  • 基准与数据集（M5竞赛、Monash时间序列、UCI）

Usage Notes

使用说明

For Claude:
  • Activate this skill for hands-on forecasting tasks, feature engineering, backtesting, or production setup
  • Start with Quick Reference and Decision Tree for fast guidance
  • Drill into references/ for detailed implementation patterns
  • Use assets/ for copy-paste ready code
  • Always check for temporal leakage (future data in training)
  • Start with strong baselines; choose model family based on horizon, covariates, and latency/cost constraints
  • Emphasize explainability for healthcare/finance domains
  • Monitor for data distribution shifts in production
Key Principle: Time series forecasting is about temporal structure, not IID assumptions. Use temporal validation, avoid future leakage, and choose models based on horizon length and data characteristics.
针对Claude：
  • 针对实操性预测任务、特征工程、回测或生产设置激活本技能
  • 从快速参考和决策树获取快速指导
  • 深入references/获取详细的实施模式
  • 使用assets/中的可直接复用代码
  • 始终检查时间数据泄露（训练中使用未来数据）
  • 从强基准模型开始；根据周期、协变量、延迟/成本约束选择模型家族
  • 针对医疗/金融领域强调可解释性
  • 生产环境中监控数据分布偏移
核心原则： 时间序列预测的核心是时间结构，而非IID假设。使用时间验证，避免未来数据泄露，根据周期长度与数据特征选择模型。