多Agent模型逐Tick运行变慢问题排查及代码性能分析

核心性能瓶颈 #

你的代码运行随时间变慢的根本原因，不是if判断的固定消耗，而是每次agent到达目标后执行的全图斑块扫描操作：

• 对于landowner，每次触发目标更新时，min-one-of patches with [Tipo-Cobertura = "ForestX"][distance myself]会遍历整个地图的所有斑块来筛选符合条件的森林斑块，再逐一计算距离找到最近的。地图规模大时，单次操作的CPU开销极高。
• 对于expansionist，min-one-of patches with [Tipo-Cobertura = "Grass" and ...]同样是全图扫描，遍历所有符合条件的草地斑块。
• 随着时间推移，要么agent数量增加（若斑块转换会生成新agent），要么符合条件的斑块分布更分散，导致每tick的全图扫描总开销持续累积，最终表现为运行越来越慢。

优化方案 #

1. 预存符合条件的斑块集合（全局缓存）

全局维护各类目标斑块的集合，仅在斑块属性变化时更新，避免agent每次都重复全图筛选：

; 全局变量定义
globals [
  forest-patches-by-type  ; 按类型存储森林斑块集合
  valid-grass-patches     ; 符合条件的草地斑块集合
]

; 初始化时构建集合
to setup
  ; ... 其他初始化逻辑
  update-target-patches
end

; 斑块属性变化时调用（比如转换斑块类型后）
to update-target-patches
  ; 更新森林斑块集合
  set forest-patches-by-type (list
    (list "Forest1" patches with [Tipo-Cobertura = "Forest1"])
    (list "Forest2" patches with [Tipo-Cobertura = "Forest2"])
    (list "Forest3" patches with [Tipo-Cobertura = "Forest3"])
    (list "Forest4" patches with [Tipo-Cobertura = "Forest4"])
    (list "Forest5" patches with [Tipo-Cobertura = "Forest5"])
  )
  ; 更新有效草地斑块集合
  set valid-grass-patches patches with [Tipo-Cobertura = "Grass" and not any? neighbors with [es-bosque = 1] and empty = 1]
end

2. 重构Landowner代码，复用缓存集合

ask landowner [
  ifelse patch-here = objetivo [
    let target-type nobody
    ; 按优先级遍历森林类型，找到邻居存在的类型
    foreach forest-patches-by-type [
      type-data ->
      let f-type first type-data
      let f-patches last type-data
      if target-type = nobody and any? neighbors with [Tipo-Cobertura = f-type] and any? f-patches [
        set objetivo min-one-of f-patches [distance myself]
        set target-type f-type
      ]
    ]
  ] [
    face objetivo fd 1
  ]
]

3. 重构Expansionist代码，复用缓存集合

ask expansionist [
  ifelse patch-here = objetivo [
    if any? neighbors with [Tipo-Cobertura = "Grass" and not any? neighbors with [es-bosque = 1]] and any? valid-grass-patches [
      set objetivo min-one-of valid-grass-patches [distance myself]
    ]
  ] [
    face objetivo fd 1
  ]
]

4. 进一步优化：限制搜索范围

如果不需要找全图最近的斑块，仅找当前位置附近的目标，可以用in-radius缩小搜索范围，大幅降低计算量：

; Landowner示例：先在半径20范围内找，找不到再 fallback 到全图
foreach forest-patches-by-type [
  type-data ->
  let f-type first type-data
  let f-patches last type-data
  if target-type = nobody and any? neighbors with [Tipo-Cobertura = f-type] and any? f-patches [
    let nearby-patches f-patches in-radius 20
    if any? nearby-patches [
      set objetivo min-one-of nearby-patches [distance myself]
    ] else [
      set objetivo min-one-of f-patches [distance myself]
    ]
    set target-type f-type
  ]
]

额外性能建议 #

• 使用NetLogo内置的profile命令，精准定位每个代码块的耗时，验证优化效果。
• 避免在ask循环中重复执行全局斑块筛选操作，尽量将集合维护放在全局层面。
• 若agent数量较多，可考虑分批处理agent，避免单tick内CPU负载过高。

内容的提问来源于stack exchange，提问作者bksnlow