🏝️
基于生物地理学的优化算法（BBO）·进阶篇
（2/3）· 当 HSI 与迁徙率接管参数寻优，多数交易者还停留在把随机搜索当万能钥匙的阶段
把优化器当成黑箱直接跑完事，是 EA 调参里最隐蔽的浪费。BBO 用岛屿宜居度给解排序、用迁入迁出交换特征，若连 HSI 和迁移率的关系都没厘清，跑出来的参数只是另一组噪声。
◍ 生物地理优化算法的结构体与初始化落点
在 MT5 里用生物地理优化（BBO）做参数寻优时，先得把栖息地（habitat）相关的状态塞进一个结构体。下面这段声明定义了每个栖息地的物种数、迁入率、迁出率和存在概率，外加全局的物种上限与精英保留数，是后续种群迭代的底层数据骨架。
MQL5 / C++
Init 函数是实盘前必须跑通的一环：它接收参数搜索区间的最小值、最大值、步长数组，以及可选的迭代轮数（默认 0 表示走类内标准轮数）。内部先调 StandardInit 做基础校验，失败直接返 false，避免脏数据进优化器。
class="type">class="kw">double emigrationMax; class=class="str">"cmt">// 最大迁出率 class="type">class="kw">double mutationProb; class=class="str">"cmt">// 变异概率 class="type">int elitismCount; class=class="str">"cmt">// 精英解数量 class="type">int speciesMax; class=class="str">"cmt">// 物种最大数量 class="kw">private: class=class="str">"cmt">//--------------------------------------------------------------- class="kw">struct S_HabitatData { class="type">int speciesCount; class=class="str">"cmt">// 栖息地内物种数 class="type">class="kw">double immigration; class=class="str">"cmt">// 迁入率 class="type">class="kw">double emigration; class=class="str">"cmt">// 迁出率 class="type">class="kw">double probability; class=class="str">"cmt">// 存在概率 }; S_HabitatData habitatData []; class=class="str">"cmt">// 每个栖息地的数据 class="type">class="kw">double probabilities []; class=class="str">"cmt">// 统计变异用的概率数组 class="type">void InitializePopulation(); class="type">void CalculateRates(); class="type">void Migration(); class="type">void Mutation(); class="type">class="kw">double CalculateProbability(class="type">int speciesCount); };
MQL5 / C++
注意 probabilities 数组长度被定为 speciesMax+1，循环把 0 到 speciesMax 每种物种数的存在概率算出来并累加，最后做归一化。若 sum 为 0（极端参数下 CalculateProbability 全返 0），归一化被跳过，调用方拿到的就是全 0 权重——这种边界情况在外汇品种点值跨度大时可能出现，建议开 MT5 把 speciesMax 调到 20 以上观察 sum 是否恒正。
class="type">bool C_AO_BBO::Init(const class="type">class="kw">double &rangeMinP [], class=class="str">"cmt">// 最小值 const class="type">class="kw">double &rangeMaxP [], class=class="str">"cmt">// 最大值 const class="type">class="kw">double &rangeStepP [], class=class="str">"cmt">// 步长 const class="type">int epochsP = class="num">0) class=class="str">"cmt">// 迭代轮数 { if (!StandardInit(rangeMinP, rangeMaxP, rangeStepP)) class="kw">return false; ArrayResize(habitatData, popSize); ArrayResize(probabilities, speciesMax + class="num">1); class="type">class="kw">double sum = class="num">0.0; for (class="type">int i = class="num">0; i <= speciesMax; i++) { probabilities [i] = CalculateProbability(i); sum += probabilities [i]; } if (sum > class="num">0) { for (class="type">int i = class="num">0; i <= speciesMax; i++) { probabilities [i] /= sum; } } class="kw">return true; }
MQL5 / C++
class="type">class="kw">double emigrationMax; class=class="str">"cmt">// maximum emigration rate class="type">class="kw">double mutationProb; class=class="str">"cmt">// mutation probability class="type">int elitismCount; class=class="str">"cmt">// number of elite solutions class="type">int speciesMax; class=class="str">"cmt">// maximum number of species class="kw">private: class=class="str">"cmt">//------------------------------------------------------------------- class="kw">struct S_HabitatData { class="type">int speciesCount; class=class="str">"cmt">// number of species in the habitat class="type">class="kw">double immigration; class=class="str">"cmt">// immigration rate class="type">class="kw">double emigration; class=class="str">"cmt">// emigration rate class="type">class="kw">double probability; class=class="str">"cmt">// probability of existence }; S_HabitatData habitatData []; class=class="str">"cmt">// data for each habitat class="type">class="kw">double probabilities []; class=class="str">"cmt">// probabilities for counting mutations class="type">void InitializePopulation(); class="type">void CalculateRates(); class="type">void Migration(); class="type">void Mutation(); class="type">class="kw">double CalculateProbability(class="type">int speciesCount); }; class="type">bool C_AO_BBO::Init(const class="type">class="kw">double &rangeMinP [], class=class="str">"cmt">// minimum values const class="type">class="kw">double &rangeMaxP [], class=class="str">"cmt">// maximum values const class="type">class="kw">double &rangeStepP [], class=class="str">"cmt">// step change const class="type">int epochsP = class="num">0) class=class="str">"cmt">// number of epochs { if (!StandardInit(rangeMinP, rangeMaxP, rangeStepP)) class="kw">return false; ArrayResize(habitatData, popSize); ArrayResize(probabilities, speciesMax + class="num">1); class="type">class="kw">double sum = class="num">0.0; for (class="type">int i = class="num">0; i <= speciesMax; i++) { probabilities [i] = CalculateProbability(i); sum += probabilities [i]; } if (sum > class="num">0) { for (class="type">int i = class="num">0; i <= speciesMax; i++) { probabilities [i] /= sum; } } class="kw">return true; }
生物地理优化在EA里的迭代骨架
把 BBO（生物地理优化）塞进 MT5 的 EA 框架，核心就是一个 Moving() 驱动单次迭代。首次进入靠 revision 标志位拦截，只做种群初始化然后直接 return，后续每次调用才跑真正的栖息地演化。
迭代主流程分五步：先用 u.Sorting 按适应度（HSI）给种群排序；再算每个栖息地的迁入迁出率；接着 Migration() 交换 SIV（候选解向量）；然后按概率 Mutation() 扰动；最后把本次目标值 f 暂存进 fP 留给下一代比对。
InitializePopulation() 里对每个 agent 的 coords 维都用 u.RNDfromCI 在 [rangeMin, rangeMax] 内均匀撒点，并立即用 u.SeInDiSp 吸附到合法步长网格——这意味着你调 rangeStep[] 会直接改变解空间的离散密度。
CalculateRates() 采用线性迁移模型：speciesCount = speciesMax - i * speciesMax / popSize，排名越靠前（i 越小）物种数越多，迁入率随之走低。排序后第 0 号栖息地物种数恒为 speciesMax，末位则为 0，这是改迁入/迁出曲线斜率的硬支点。
别把正态当圣经：BBO 的物种数分布是确定性线性递减，不是随机高斯。若你之前用 GA 的突变逻辑套过来，迁入率公式会直接算错，建议在 MT5 里打印 habitatData[i].speciesCount 验证前几代数值。
MQL5 / C++
class="type">void C_AO_BBO::Moving() { class=class="str">"cmt">// First iteration - initialization of the initial population if (!revision) { InitializePopulation(); revision = true; class="kw">return; } class=class="str">"cmt">// Main optimization class=class="str">"cmt">// class="num">1. Sort the population by HSI(fitness) class="kw">static S_AO_Agent aTemp []; ArrayResize(aTemp, popSize); u.Sorting(a, aTemp, popSize); class=class="str">"cmt">// class="num">2. Calculate immigration and emigration rates CalculateRates(); class=class="str">"cmt">// class="num">3. Migration(exchange of SIVs between habitats) Migration(); class=class="str">"cmt">// class="num">4. Probability-based mutation Mutation(); class=class="str">"cmt">// class="num">5. Save state for the next iteration for (class="type">int i = class="num">0; i < popSize; i++) { a [i].fP = a [i].f; } } class="type">void C_AO_BBO::Revision() { class=class="str">"cmt">// Find the best solution in the current population for (class="type">int i = class="num">0; i < popSize; i++) { class=class="str">"cmt">// Update the best solution if (a [i].f > fB) { fB = a [i].f; ArrayCopy(cB, a [i].c, class="num">0, class="num">0, WHOLE_ARRAY); } } } class="type">void C_AO_BBO::InitializePopulation() { class=class="str">"cmt">// Initialize the initial population uniformly throughout the space for (class="type">int i = class="num">0; i < popSize; i++) { for (class="type">int c = class="num">0; c < coords; c++) { class=class="str">"cmt">// Generate random coordinates within acceptable limits a [i].c [c] = u.RNDfromCI(rangeMin [c], rangeMax [c]); class=class="str">"cmt">// Round to the nearest acceptable step a [i].c [c] = u.SeInDiSp(a [i].c [c], rangeMin [c], rangeMax [c], rangeStep [c]); } class=class="str">"cmt">// Initialize habitat data habitatData [i].speciesCount = class="num">0; habitatData [i].immigration = class="num">0.0; habitatData [i].emigration = class="num">0.0; habitatData [i].probability = class="num">0.0; } } class="type">void C_AO_BBO::CalculateRates() { class=class="str">"cmt">// For the linear migration model for (class="type">int i = class="num">0; i < popSize; i++) { class=class="str">"cmt">// The number of species is inversely proportional to the rank(the best solutions have more species) habitatData [i].speciesCount = speciesMax - (i * speciesMax / popSize); class=class="str">"cmt">// The rate of immigration decreases as the number of species increases
「生物地理迁移算子的代码落点」
这套优化器的栖息地间迁移，核心是把「迁入率」和「迁出率」当概率闸门用。迁入率随物种数逼近上限而衰减，公式是 immigrationMax * (1.0 - speciesCount/speciesMax)；迁出率则正比于物种占比，emigrationMax * speciesCount/speciesMax。当某栖息地物种数超过 speciesMax，其存在概率直接置 0，相当于被算法淘汰。 Migration() 函数里，elitismCount 之前的精英解被 continue 跳过，不参与变异，保证最优结构不丢。对每个非精英栖息地，先以 habitatData[i].immigration 为阈值扔随机数，命中才进入逐坐标（SIV）修改；每个坐标又独立以同样迁入率为概率，决定是否真的从该栖息地的某一维抄值。 源栖息地不是随机挑的，而是按所有其他栖息地的迁出率做轮盘赌。代码先累加 sumEmigration，再掷出 roulette = RNDprobab() * sumEmigration，线性扫描累加到 cumSum 首次越过 roulette 的 j，就把 a[j].c[c] 抄给 a[i].c[c]。这一层双重概率判断，使高迁出率且非自身的栖息地更可能被选为供体，低物种数栖息地的维度被改写概率也更高。 在 MT5 里把这段直接塞进 EA 的 C_AO_BBO 类，调小 immigrationMax 会明显压低整体扰动频率，回测中种群收敛可能更慢但更稳；外汇与贵金属品种波动跳跃大，直接挂实盘属高风险，建议先用历史数据跑种群分布。
MQL5 / C++
habitatData [i].immigration = immigrationMax * (class="num">1.0 - (class="type">class="kw">double)habitatData [i].speciesCount / speciesMax); class=class="str">"cmt">// The rate of emigration increases with the number of species habitatData [i].emigration = emigrationMax * (class="type">class="kw">double)habitatData [i].speciesCount / speciesMax; class=class="str">"cmt">// Probability of habitat existence if (habitatData [i].speciesCount <= speciesMax) { habitatData [i].probability = probabilities [habitatData [i].speciesCount]; } else { habitatData [i].probability = class="num">0.0; } } class=class="str">"cmt">//—————————————————————————————————————————————————————————————————————————————— class=class="str">"cmt">//+----------------------------------------------------------------------------+ class=class="str">"cmt">//| Migration(exchange of SIVs between habitats) | class=class="str">"cmt">//+----------------------------------------------------------------------------+ class="type">void C_AO_BBO::Migration() { for (class="type">int i = class="num">0; i < popSize; i++) { class=class="str">"cmt">// Skip elite solutions if (i < elitismCount) class="kw">continue; class=class="str">"cmt">// Determine whether the habitat will be modified if (u.RNDprobab() < habitatData [i].immigration) { class=class="str">"cmt">// For each coordinate(SIV) for (class="type">int c = class="num">0; c < coords; c++) { class=class="str">"cmt">// Determine whether this coordinate will be modified if (u.RNDprobab() < habitatData [i].immigration) { class=class="str">"cmt">// Select a migration source based on emigration rates class="type">class="kw">double sumEmigration = class="num">0.0; for (class="type">int j = class="num">0; j < popSize; j++) { if (j != i) sumEmigration += habitatData [j].emigration; } if (sumEmigration > class="num">0) { class=class="str">"cmt">// Roulette source selection class="type">class="kw">double roulette = u.RNDprobab() * sumEmigration; class="type">class="kw">double cumSum = class="num">0.0; for (class="type">int j = class="num">0; j < popSize; j++) { if (j != i) { cumSum += habitatData [j].emigration; if (roulette <= cumSum) { class=class="str">"cmt">// Copy SIV from habitat j to habitat i a [i].c [c] = a [j].c [c]; break; } } } } } } } } }
◍ 按生存概率反向调变异率
生物地理优化里，精英解不参与变异，直接保留进下一代；其余个体才进入扰动流程。这段代码把变异概率和该个体的存在概率挂钩：存在概率越高，变异率越低，公式写成 mutationRate = mutationProb * (1.0 - habitatData[i].probability)。 具体实现里，先用 u.RNDprobab() 抽一个随机数，小于计算出的 mutationRate 才触发变异；变异坐标用 MathRand() % coords 随机挑一个维度，新值从对应维度的区间里取，再用 SeInDiSp 对齐到步长网格。这样高适应度个体倾向稳定，低适应度个体更容易被重采样。 CalculateProbability 用了一个简化模型：物种数落在 speciesMax/2 附近时概率最高，偏离均衡点按高斯型衰减，返回 MathExp(-d²/(2·eq²))。若 speciesMax 设 100，均衡点就是 50，物种数偏离到 80 时概率约降到 0.24，偏离越远越接近 0。 在 MT5 里把这段接进自己的 EA 框架时，重点盯 mutationProb 和 rangeStep：步长过粗会让变异跳不到精细参数区，过细又拖慢收敛。外汇与贵金属杠杆高、滑点随机，任何优化结果都只是概率倾向，实盘前请用历史数据多周期回测验证。
MQL5 / C++
class="type">void C_AO_BBO::Mutation() { for (class="type">int i = class="num">0; i < popSize; i++) { class=class="str">"cmt">// Skip elite solutions if (i < elitismCount) class="kw">continue; class=class="str">"cmt">// The mutation rate is inversely proportional to the probability of existence class="type">class="kw">double mutationRate = mutationProb * (class="num">1.0 - habitatData [i].probability); if (u.RNDprobab() < mutationRate) { class=class="str">"cmt">// Select a random coordinate for mutation class="type">int mutateCoord = MathRand() % coords; class=class="str">"cmt">// Generate a new value for the selected coordinate a [i].c [mutateCoord] = u.RNDfromCI(rangeMin [mutateCoord], rangeMax [mutateCoord]); a [i].c [mutateCoord] = u.SeInDiSp(a [i].c [mutateCoord], rangeMin [mutateCoord], rangeMax [mutateCoord], rangeStep [mutateCoord]); } } } class="type">class="kw">double C_AO_BBO::CalculateProbability(class="type">int speciesCount) { class=class="str">"cmt">// Simplified probability model class=class="str">"cmt">// Maximum probability in the middle of the range(equilibrium) class="type">int equilibrium = speciesMax / class="num">2; class="type">class="kw">double distance = MathAbs(speciesCount - equilibrium); class="type">class="kw">double probability = MathExp(-distance * distance / (class="num">2.0 * equilibrium * equilibrium)); class="kw">return probability; }
让小布替你跑这套群岛模拟
这些诊断小布盯盘的 AIGC 已内置，打开对应品种页即可看到迁徙率与适宜度指数的实时映射，把重复劳动交给小布，你专注决策。
常见问题
HSI 直接对应岛屿宜居程度即解质量，且迁移率由物种分布模型推导，优质解主动共享特征而稳定性强，不像适应度仅做选择权重。
小布盯盘目前以内置诊断与看板为主，BBO 类优化需在 MQL5 策略测试器自行部署，但小布可辅助你监控迁移收敛过程。
劣质解物种少、空间大，高迁入率使其持续吸收优质解特征，避免种群过早陷在同一峰区，提升跳出局部最优的概率。
规模与评估函数复杂度线性相关，50 岛属轻量设置，若目标函数含大量指标调用建议分批迁移以降低单次开销。