RAILWISE-CLI 智能体配置详解

RAILWISE-CLI 智能体配置详解

目标读者：需要自定义智能体行为的高级用户、二次开发人员
预计阅读时间：20 分钟
前置要求：已完成 cli-quickstart.md 快速入门

---

1. 智能体架构概览

RAILWISE-CLI 采用 多智能体协作架构（Multi-Agent Collaboration），每个智能体是独立的任务执行单元，通过消息总线进行协作。系统内置四大核心智能体：

智能体	标识符	职责	典型任务
测量智能体	surveyor	原始数据解析与预处理	数据导入、坐标转换、观测值提取
平差智能体	adjuster	测量平差与精度评定	间接平差、条件平差、精度分析
监测智能体	monitor	变形监测与预警分析	时序分析、趋势预测、预警分级
质检智能体	inspector	数据质量检查与合规校验	限差检查、粗差探测、规范符合性

1.1 智能体生命周期

初始化 (init) → 配置加载 (config) → 任务接收 (receive)
→ 执行处理 (process) → 结果输出 (output) → 日志归档 (archive)

---

2. 内置智能体配置

2.1 测量智能体（surveyor）

配置文件路径：.railwise/agents/surveyor.yaml

agent:
  name: "surveyor"
  version: "1.2.0"
  description: "原始测量数据解析与预处理智能体"

# 支持的仪器品牌与型号
instruments:
  leica:
    models: ["TS16", "TS60", "LS15", "LS10"]
    formats: ["gsi", "gsi16", "fiducial"]
    default_encoding: "utf-8"
  topcon:
    models: ["GT-1200", "GM-52"]
    formats: ["topcon-raw", "sdr"]
  south:
    models: ["NTS-362R", "NTS-372R"]
    formats: ["south-raw"]
  trimble:
    models: ["S9", "R12"]
    formats: ["trimble-dc", "rinex"]

# 坐标系统配置
coordinate_systems:
  default: "CGCS2000"
  supported:
    - name: "CGCS2000"
      epsg: 4490
      projection: "高斯-克吕格 3度带"
    - name: "宁波2000"
      epsg: 4547
      local_params:
        central_meridian: 121.0
        false_easting: 500000
        false_northing: 0

# 数据预处理规则
preprocessing:
  auto_detect_encoding: true
  skip_invalid_records: false      # 遇到错误记录时：false=报错，true=跳过
  fill_missing_station_names: true
  default_instrument_height: 1.600   # 默认仪器高（米）
  default_prism_height: 1.500        # 默认棱镜高（米）

如需添加地方独立坐标系，请在 local_params 中定义中央子午线、假东、假北等参数。支持七参数和四参数转换。

2.2 平差智能体（adjuster）

配置文件路径：.railwise/agents/adjuster.yaml

agent:
  name: "adjuster"
  version: "1.2.0"
  description: "测量平差与精度评定智能体"

# 平差方法配置
adjustment_methods:
  indirect:
    name: "间接平差"
    suitable_for: ["水准网", "导线网", "GPS网"]
    solver: "最小二乘法"
    max_iterations: 100
    convergence_threshold: 1.0e-6

  conditional:
    name: "条件平差"
    suitable_for: ["单一水准路线", "闭合导线"]
    solver: "最小二乘法"

  robust:
    name: "稳健平差"
    suitable_for: ["含粗差数据"]
    solver: "Huber法"
    tuning_constant: 1.345

# 定权方案
weight_schemes:
  by_distance:
    name: "按距离定权"
    formula: "P = C / S"           # C 为常数，S 为测站距离（km）
    default_c: 100
  by_observations:
    name: "按测回数定权"
    formula: "P = n"               # n 为测回数
  equal:
    name: "等权"
    formula: "P = 1"

# 精度评定
precision_evaluation:
  compute_unit_variance: true
  compute_point_errors: true
  compute_relative_errors: true
  confidence_level: 0.95
  error_ellipse_scale: 1.0          # 误差椭圆缩放系数

2.3 监测智能体（monitor）

配置文件路径：.railwise/agents/monitor.yaml

agent:
  name: "monitor"
  version: "1.2.0"
  description: "变形监测与预警分析智能体"

# 监测类型配置
monitor_types:
  settlement:
    name: "沉降监测"
    unit: "mm"
    warning_thresholds: [2, 4, 6]     # 黄、橙、红三级预警（mm/次）
    cumulative_limit: 30               # 累计沉降限值（mm）
  horizontal:
    name: "水平位移"
    unit: "mm"
    warning_thresholds: [3, 5, 8]
  convergence:
    name: "收敛监测"
    unit: "mm"
    warning_thresholds: [5, 10, 15]
  inclinometer:
    name: "测斜"
    unit: "mm/m"
    warning_thresholds: [2, 4, 6]

# 预警方法
alert_methods:
  standard:
    name: "规范法"
    description: "依据 GB 50497 / JGJ 8 等规范阈值"
  statistical:
    name: "统计法"
    description: "基于历史数据的 3σ 原则"
    min_history_points: 10
  combined:
    name: "综合法"
    description: "规范法与统计法取严格者"

# 时序分析
time_series:
  trend_model: "线性回归"              # 可选：线性回归 / 多项式 / 指数平滑
  prediction_horizon: 7               # 预测未来 7 期
  seasonal_adjustment: false

# 报告输出
reporting:
  include_trend_chart: true
  include_prediction: true
  alert_summary_format: "表格+文字"

规范法阈值依据 GB 50497-2019 附录取值，但实际工程中需结合地质条件、支护形式、周边环境保护等级综合确定。修改阈值前请咨询项目技术负责人。

2.4 质检智能体（inspector）

配置文件路径：.railwise/agents/inspector.yaml

agent:
  name: "inspector"
  version: "1.2.0"
  description: "数据质量检查与合规校验智能体"

# 检查项配置
check_items:
  # 限差检查
  tolerance:
    enabled: true
    checks:
      - name: "水准测量限差"
        standard: "GB 50026-2020"
        category: "二等水准"
        items: ["前后视距差", "累积视距差", "红黑面读数差", "高差之差"]
      - name: "导线测量限差"
        standard: "GB 50026-2020"
        category: "一级导线"
        items: ["测角中误差", "方位角闭合差", "导线全长相对闭合差"]

  # 粗差探测
  outlier_detection:
    enabled: true
    methods: ["Baarda法", "Tau检验"]
    significance_level: 0.05
    min_redundancy: 1

  # 规范符合性
  compliance:
    enabled: true
    standards: ["GB 50026", "GB 50497", "GB 50911", "JGJ 8"]
    check_items: ["观测频次", "精度等级", "仪器标称精度"]

# 输出配置
output:
  format: "detailed"                  # detailed / summary / json
  fail_fast: false                    # true=发现首个错误即停止
  generate_correction_suggestions: true

---

3. 自定义智能体开发

3.1 创建自定义智能体

在 .railwise/agents/ 目录下创建新的 YAML 配置文件：

.railwise/agents/custom-uav.yaml

agent:
  name: "uav-survey"
  version: "1.0.0"
  description: "无人机航测数据处理智能体"
  extends: "surveyor"               # 继承内置智能体能力

# 自定义参数
capabilities:
  - "photo_alignment"
  - "point_cloud_generation"
  - "dsm_creation"
  - "orthomosaic_export"

supported_formats:
  - "exif"
  - "geotiff"
  - "las"
  - "ply"

processing:
  ground_sample_distance: 0.02        # 地面采样距离（米）
  overlap_forward: 80                 # 航向重叠度（%）
  overlap_side: 70                    # 旁向重叠度（%）

3.2 注册与使用

# 注册自定义智能体
railwise agent register .railwise/agents/custom-uav.yaml

# 在工作流中引用
```yaml
steps:
  - name: "无人机数据处理"
    agent: "uav-survey"
    action: "process"
    input: "data/uav/flight_001/"

---

4. 智能体间通信与协作

4.1 消息格式

智能体通过标准消息格式交换数据：

interface AgentMessage {
  id: string;                        // 消息唯一标识
  from: string;                      // 发送方智能体名称
  to: string;                        // 接收方智能体名称（"*" 表示广播）
  type: "data" | "command" | "event" | "error";
  payload: unknown;                  // 消息载荷
  timestamp: string;                 // ISO 8601 时间戳
  metadata: {
    workflowId: string;
    stepIndex: number;
    priority: "low" | "normal" | "high" | "critical";
  };
}

4.2 工作流中的数据传递

steps:
  - name: "数据导入"
    agent: "surveyor"
    action: "import"
    output_var: "raw_data"            # 将结果存入变量

  - name: "数据质检"
    agent: "inspector"
    action: "check"
    input_var: "raw_data"             # 引用上一步变量
    output_var: "check_result"

  - name: "条件执行平差"
    agent: "adjuster"
    action: "adjust"
    input_var: "raw_data"
    condition: "check_result.passed"  # 条件执行

---

5. 性能调优

5.1 并发控制

.railwise/config.yaml

performance:
  max_concurrent_agents: 4          # 最大并发智能体数
  memory_limit_per_agent: "512MB"
  enable_parallel_processing: true

5.2 缓存策略

cache:
  enabled: true
  directory: ".railwise/cache"
  ttl: 86400                         # 缓存有效期（秒）
  max_size: "2GB"

---

6. 相关文档

• cli-quickstart.md — 快速入门指南
• cli-workflow-orchestration.md — 工作流编排
• cli-data-formats.md — 数据格式支持
• cli-troubleshooting.md — 常见问题排查

---

元数据标签