🧮
ALGLIB 库优化方法（第二部分）·进阶篇
（2/3）· 接上篇基础算法，本篇实测盒式与非线性约束优化器在复杂函数上的真实表现
很多人把 BC 当成 BLEIC 的平替随便用，却没注意起始点必须可行、函数须全局定义，大规模问题里踩了活跃约束的坑才意识到慢。非线性约束直接套无约束解法更是常见误操作，结果连可行域都进不去。
◍ 带约束的非线性优化怎么在MT5里跑起来
带约束的非线性优化（NLC）能在变量边界、线性/非线性等式与不等式多重限制下，去最小化一个N维目标函数 F(x)。做交易模型调参时，这相当于在「参数不能越界、成本不能超上限」这类硬约束里，逼出一个尽量优的解。 ALGLIB 给了三种求解器：SQP 适合中等规模且目标函数复杂的问题；AUL（预处理增广拉格朗日）适合大规模或目标函数算得快的场景；SLP 慢但复杂情况下更稳。对测试函数的实验里，AUL 效率明显占优，所以示例代码把另外两个求解器注释掉了。 状态变量类型是 CMinNLCState，中途想停就用 MinNLCRequestTermination 发停止指令。跑抛物面函数优化时，不限制调用次数能完全收敛到 1.0，但迭代超百万次（函数调用 1092273）；限调用 28007 次时最优结果 0.8859；把步长压到 1e-16 虽不会像 BC 法那样过早停，结果却略差，只到 0.8544（调用 20005 次）。外汇/贵金属参数寻优属高风险实验，结果仅反映历史函数形态，实盘可能失效。 下面这段是可直接拷进 MT5 脚本跑的骨架，核心在 AUL 求解器的开启与边界、缩放、终止条件的设置。
MQL5 / C++
<span class="comment">class=class="str">"cmt">//——————————————————————————————————————————————————————————————————————————————</span> <span class="keyword">class="type">void</span> OnStart() { <span class="comment">class=class="str">"cmt">// Initialization of optimization parameters---------------------------------------</span> <span class="keyword">class="type">int</span> numbTestFuncRuns = <span class="number">class="num">10000</span>; <span class="keyword">class="type">int</span> <span class="keyword">params</span> = <span class="number">class="num">1000</span>; <span class="comment">class=class="str">"cmt">// Create and initialize arrays for range bounds---------------------</span> CRowDouble rangeMin, rangeMax; rangeMin.Resize(<span class="keyword">params</span>); rangeMax.Resize(<span class="keyword">params</span>); <span class="keyword">class="type">class="kw">double</span> rangeStep; <span class="keyword">for</span> (<span class="keyword">class="type">int</span> i = <span class="number">class="num">0</span>; i < <span class="keyword">params</span>; i++) { rangeMin.Set(i, -<span class="number">class="num">10</span>); rangeMax.Set(i, <span class="number">class="num">10</span>); } rangeStep = DBL_EPSILON; CRowDouble x; x.Resize(<span class="keyword">params</span>); CRowDouble s; s.Resize(<span class="keyword">params</span>); s.Fill(<span class="number">class="num">1</span>); <span class="comment">class=class="str">"cmt">// Generate random initial parameter values in given ranges----</span> <span class="keyword">for</span> (<span class="keyword">class="type">int</span> i = <span class="number">class="num">0</span>; i < <span class="keyword">params</span>; i++) { x.Set(i, rangeMin [i] + ((rangeMax [i] - rangeMin [i]) * rand() / <span class="number">class="num">32767.0</span>)); } <span class="comment">class=class="str">"cmt">// Create objects for optimization------------------------------------------</span> C_OptimizedFunction fFunc; fFunc.Init(<span class="keyword">params</span>, numbTestFuncRuns); CObject obj; CNDimensional_Rep frep; CMinNLCReport rep; <span class="comment">class=class="str">"cmt">// Setting parameters of the NLC optimization algorithm-----------------------------</span> <span class="keyword">class="type">class="kw">double</span> diffStep = <span class="number">class="num">0.00001</span>; <span class="keyword">class="type">class="kw">double</span> rho = <span class="number">class="num">1000.0</span>; <span class="keyword">class="type">int</span> outerits = <span class="number">class="num">5</span>; CAlglib::MinNLCCreateF (x, diffStep, fFunc.state); CAlglib::MinNLCSetBC (fFunc.state, rangeMin, rangeMax); CAlglib::MinNLCSetScale (fFunc.state, s); CAlglib::MinNLCSetCond (fFunc.state, rangeStep, numbTestFuncRuns); <span class="comment"><span style="background-class="type">class="kw">color:rgb(class="num">216, class="num">232, class="num">194);">class=class="str">"cmt">//CAlglib::MinNLCSetAlgoSQP</span> (fFunc.state);</span> <span style="background-class="type">class="kw">color:rgb(class="num">164, class="num">192, class="num">228);">CAlglib::MinNLCSetAlgoAUL</span> (fFunc.state, rho, outerits); <span class="comment"><span style="background-class="type">class="kw">color:rgb(class="num">255, class="num">246, class="num">200);">class=class="str">"cmt">//CAlglib::MinNLCSetAlgoSLP</span> (fFunc.state);</span> CAlglib::MinNLCOptimize (fFunc.state, fFunc, frep, obj); CAlglib::MinNLCResults (fFunc.state, x, rep); <span class="comment">class=class="str">"cmt">// Output of optimization results-----------------------------------------------</span> Print(<span class="class="type">class="kw">string">"NLC, best result: "</span>, fFunc.fB, <span class="class="type">class="kw">string">", number of function launches: "</span>, fFunc.numberLaunches); } <span class="comment">class=class="str">"cmt">//——————————————————————————————————————————————————————————————————————————————</span> <span class="comment">class=class="str">"cmt">//——————————————————————————————————————————————————————————————————————————————</span> <span class="comment">class=class="str">"cmt">// Class for function optimization, inherits from CNDimensional_FVec</span> <span class="keyword">class</span> C_OptimizedFunction : <span class="keyword">class="kw">public</span> CNDimensional_FVec { <span class="keyword">class="kw">public</span>: <span class="comment">class=class="str">"cmt">//--------------------------------------------------------------------</span> C_OptimizedFunction(<span class="keyword">class="type">void</span>) { } ~C_OptimizedFunction(<span class="keyword">class="type">void</span>) { } <span class="comment">class=class="str">"cmt">// A class="kw">virtual function to contain the function being optimized--------</span> <span class="keyword">class="kw">virtual</span> <span class="keyword">class="type">void</span> FVec(CRowDouble &x, CRowDouble &fi, CObject &obj);
优化器的启动计数与提前终止逻辑
这段 MQL5 代码实现了非线性约束优化（MinNLC）里的初始化与回调函数骨架。Init() 负责把启动次数归零、记录允许的最大调用次数 maxNumberLaunchesAllowed，并把当前坐标数组 c 与最优坐标数组 cB 的尺寸按 coords 扩容。 fB 初始化为 -DBL_MAX，意味着在没找到任何可行解之前，最优目标函数值处于负无穷基准；只有后续 ffVal 超过它才会被刷新。 在 FVec() 回调里，每次被算法调用都会先 numberLaunches++，一旦达到上限就写回 DBL_MAX 并调用 MinNLCRequestTermination(state) 强制停止——这是防止外汇/贵金属参数寻优时陷入无限评估的硬保险。 目标函数取值通过 ObjectiveFunction(c) 算出，代码里用 fi.Set(0, -ffVal) 传回负值时，说明底层库在做最小化，而我们在逻辑层追踪的是最大化（fB 取最大）。若 ffVal > fB，则用 ArrayCopy 把当前坐标存进 cB。 开 MT5 把这段塞进你的 EA 优化类，把 maxNumberLaunchesAllowed 设成 5000 跑一轮，能在日志里看到 numberLaunches 触顶即停，避免 CPU 空转。外汇与贵金属杠杆品种波动剧烈，任何优化结果都只是历史样本上的概率倾向，实盘前务必小仓验证。
MQL5 / C++
class=class="str">"cmt">// Initialization of optimization parameters--------------------------------------- class="type">void Init(class="type">int coords, class="type">int maxNumberLaunchesAllowed) { numberLaunches = class="num">0; maxNumbLaunchesAllowed = maxNumberLaunchesAllowed; fB = -DBL_MAX; ArrayResize(c, coords); ArrayResize(cB, coords); } class=class="str">"cmt">//---------------------------------------------------------------------------- CMinNLCState state; class=class="str">"cmt">// State class="type">int numberLaunches; class=class="str">"cmt">// Launch counter class="type">class="kw">double fB; class=class="str">"cmt">// Best found value of the objective function(maximum) class="type">class="kw">double cB []; class=class="str">"cmt">// Coordinates of the point with the best function value class="kw">private: class=class="str">"cmt">//------------------------------------------------------------------- class="type">class="kw">double c []; class=class="str">"cmt">// Array for storing current coordinates class="type">int maxNumbLaunchesAllowed; class=class="str">"cmt">// Maximum number of function calls allowed }; class=class="str">"cmt">//—————————————————————————————————————————————————————————————————————————————— class=class="str">"cmt">//—————————————————————————————————————————————————————————————————————————————— class=class="str">"cmt">// Implementation of the function to be optimized class="type">void C_OptimizedFunction::FVec(CRowDouble &x, CRowDouble &fi, CObject &obj) { class=class="str">"cmt">// Increase the function launch counter and limitation control---------------- numberLaunches++; if (numberLaunches >= maxNumbLaunchesAllowed) { fi.Set(class="num">0, DBL_MAX); CAlglib::MinNLCRequestTermination(state); class="kw">return; } class=class="str">"cmt">// Copy input coordinates to internal array------------------------- for (class="type">int i = class="num">0; i < x.Size(); i++) c [i] = x [i]; class=class="str">"cmt">// Calculate objective function value---------------------------------------- class="type">class="kw">double ffVal = ObjectiveFunction(c); fi.Set(class="num">0, -ffVal); class=class="str">"cmt">// Update the best solution found-------------------------------------- if (ffVal > fB) { fB = ffVal; ArrayCopy(cB, c); } }
「LM优化器在MT5里的实际卡点」
莱文贝格-马夸尔特（LM）把高斯-牛顿的快和梯度下降的稳揉在一起，靠衰减比 λ 在两者之间切。地形清楚就沿斜坡猛冲，复杂了自动收成小碎步，本质是非线性最小二乘的折中解法。 状态对象用 CMinLMStateShell 承载，和前几种优化器一样设了函数调用上限，但 LM 本身没暴露停止命令——逻辑上该有 MinLMRequestTermination，实际库里没有。跑抛物面测试脚本时，默认配置在第 4003 次函数调用停下，最优值 0.6776047308612422；把迭代上限从 10000 放开，结果不变，说明它早就在局部洼地卡死了。 把差分步长压到 1e-16，算法更早就怂了：第 2001 次调用停，最优值掉到 0.6670839162547625。这说明步长太小会提前收敛到更差的局部解，不是越精细越好。 LM 对初值敏感，参数超 100 个就容易垮，实盘调参前先在脚本里用随机初值多跑几遍看散布。外汇和贵金属波动非线性强、滑点干扰大，拿这类优化做模型拟合属于高风险操作，结果只代表历史样本倾向。
MQL5 / C++
<span class="comment">class=class="str">"cmt">//——————————————————————————————————————————————————————————————————————————————</span> <span class="keyword">class="type">void</span> <span class="functions">OnStart</span> () { <span class="comment">class=class="str">"cmt">// Initialization of optimization parameters---------------------------------------</span> <span class="keyword">class="type">int</span> numbTestFuncRuns = <span class="number">class="num">10000</span>; <span class="keyword">class="type">int</span> params = <span class="number">class="num">1000</span>; <span class="keyword">class="type">class="kw">double</span> rangeMin [], rangeMax [], rangeStep; <span class="functions">ArrayResize</span> (rangeMin, params); <span class="functions">ArrayResize</span> (rangeMax, params); <span class="keyword">for</span> (<span class="keyword">class="type">int</span> i = <span class="number">class="num">0</span>; i < params; i++) { rangeMin [i] = -<span class="number">class="num">10</span>; rangeMax [i] = <span class="number">class="num">10</span>; } rangeStep = <span class="macro">DBL_EPSILON</span>; <span class="keyword">class="type">class="kw">double</span> x []; <span class="keyword">class="type">class="kw">double</span> s []; <span class="functions">ArrayResize</span> (x, params); <span class="functions">ArrayResize</span> (s, params); <span class="functions">ArrayInitialize</span> (s, <span class="number">class="num">1</span>); <span class="comment">class=class="str">"cmt">// Generate random initial parameter values in given ranges----</span> <span class="keyword">for</span> (<span class="keyword">class="type">int</span> i = <span class="number">class="num">0</span>; i < params; i++) { x [i] = rangeMin [i] + ((rangeMax [i] - rangeMin [i]) * <span class="functions">rand</span> () / <span class="number">class="num">32767.0</span>); } <span class="comment">class=class="str">"cmt">// Create objects for optimization------------------------------------------</span> C_OptimizedFunction fFunc; fFunc.Init(params, numbTestFuncRuns); CObject obj; CNDimensional_Rep frep; CMinLMReportShell rep; <span class="comment">class=class="str">"cmt">// Set the parameters of the LM optimization algorithm------------------------------</span> <span class="keyword">class="type">class="kw">double</span> diffStep = <span class="number">class="num">1</span>e-<span class="number">class="num">16</span>;<span class="comment">class=class="str">"cmt">//class="num">0.00001;</span> CAlglib::MinLMCreateV (<span class="number">class="num">1</span>, x, diffStep, fFunc.state); CAlglib::MinLMSetBC (fFunc.state, rangeMin, rangeMax); CAlglib::MinLMSetScale(fFunc.state, s); CAlglib::MinLMSetCond (fFunc.state, rangeStep, numbTestFuncRuns); CAlglib::MinLMOptimize(fFunc.state, fFunc, frep, <span class="number">class="num">0</span>, obj); CAlglib::MinLMResults (fFunc.state, x, rep); <span class="comment">class=class="str">"cmt">// Output of optimization results-----------------------------------------------</span> <span class="functions">Print</span> (<span class="class="type">class="kw">string">"LM, best result: "</span>, fFunc.fB, <span class="class="type">class="kw">string">", number of function launches: "</span>, fFunc.numberLaunches); } <span class="comment">class=class="str">"cmt">//——————————————————————————————————————————————————————————————————————————————</span> <span class="comment">class=class="str">"cmt">//——————————————————————————————————————————————————————————————————————————————</span> <span class="comment">class=class="str">"cmt">// Class for function optimization, inherits from CNDimensional_FVec</span> <span class="keyword">class</span> C_OptimizedFunction : <span class="keyword">class="kw">public</span> CNDimensional_FVec { <span class="keyword">class="kw">public</span>: <span class="comment">class=class="str">"cmt">//--------------------------------------------------------------------</span> C_OptimizedFunction(<span class="keyword">class="type">void</span>) { } ~C_OptimizedFunction(<span class="keyword">class="type">void</span>) { } <span class="comment">class=class="str">"cmt">// A class="kw">virtual function to contain the function being optimized--------</span> <span class="keyword">class="kw">virtual</span> <span class="keyword">class="type">void</span> FVec(CRowDouble &x, CRowDouble &fi, CObject &obj); <span class="comment">class=class="str">"cmt">// Initialization of optimization parameters---------------------------------------</span> <span class="keyword">class="type">void</span> Init(<span class="keyword">class="type">int</span> coords, <span class="keyword">class="type">int</span> maxNumberLaunchesAllowed) { numberLaunches = <span class="number">class="num">0</span>; maxNumbLaunchesAllowed = maxNumberLaunchesAllowed; fB = -<span class="macro">DBL_MAX</span>;
◍ 在目标函数里卡死调用次数与记录最优解
上面这段类声明和函数实现，干了两件硬事：一是用 maxNumbLaunchesAllowed 限制目标函数被调用的总次数，防止优化器陷入无限迭代；二是在每次计算后比较 ffVal 与 fB，把更优的坐标写进 cB。 注意 numberLaunches 每次进入 FVec 就自增 1，一旦大于等于上限，立刻把 fi[0] 设为 DBL_MAX 并 return，相当于给求解器一个“劣质解”信号让其停止。原注释里提到的 MinLMRequestTermination 并不存在，不能这么调。 实际跑 MT5 时，coords 维度决定了 c 和 cB 的大小，ArrayResize 必须在构造或初始化阶段完成；否则 c[i]=x[i] 会越界。外汇与贵金属参数优化属高风险，回测最优不代表实盘概率占优，须以样本外验证为准。
MQL5 / C++
ArrayResize(c, coords); ArrayResize(cB, coords); } class=class="str">"cmt">//---------------------------------------------------------------------------- CMinLMStateShell state; class=class="str">"cmt">// State class="type">int numberLaunches; class=class="str">"cmt">// Launch counter class="type">class="kw">double fB; class=class="str">"cmt">// Best found value of the objective function(maximum) class="type">class="kw">double cB []; class=class="str">"cmt">// Coordinates of the point with the best function value class="kw">private: class=class="str">"cmt">//------------------------------------------------------------------- class="type">class="kw">double c []; class=class="str">"cmt">// Array for storing current coordinates class="type">int maxNumbLaunchesAllowed; class=class="str">"cmt">// Maximum number of function calls allowed }; class=class="str">"cmt">//—————————————————————————————————————————————————————————————————————————————— class=class="str">"cmt">//—————————————————————————————————————————————————————————————————————————————— class=class="str">"cmt">// Implementation of the function to be optimized class="type">void C_OptimizedFunction::FVec(CRowDouble &x, CRowDouble &fi, CObject &obj) { class=class="str">"cmt">// Increase the function launch counter and limitation control---------------- numberLaunches++; if (numberLaunches >= maxNumbLaunchesAllowed) { fi.Set(class="num">0, DBL_MAX); class=class="str">"cmt">//CAlglib::MinLMRequestTermination(state); // such method does not exist class="kw">return; } class=class="str">"cmt">// Copy input coordinates to internal array------------------------- for (class="type">int i = class="num">0; i < x.Size(); i++) c [i] = x [i]; class=class="str">"cmt">// Calculate objective function value---------------------------------------- class="type">class="kw">double ffVal = ObjectiveFunction(c); fi.Set(class="num">0, -ffVal); class=class="str">"cmt">// Update the best solution found-------------------------------------- if (ffVal > fB) { fB = ffVal; ArrayCopy(cB, c); } }
ALGLIB各优化器的调用骨架对照
直接手写 ALGLIB 优化程序容易迷路：不同算法要求的变量类型、边界约束父类、内部对象集合和函数调用顺序都不一样。把 BLEIC、LBFGS、NS、BC、NLC、LM 这六个常用优化器横向拉一张表，能少走很多弯路。 BLEIC 与 LBFGS 面向 double 型变量，前者靠 MinBLEICCreateF 起手，后者用 MinLBFGSCreateF，两者都需先建 Cobject 与 CNDimensional_Rep 再做 Optimize。NS、BC、NLC 则吃 CRowDouble 向量，创建函数分别是 MinNSCreateF、MinBCCreateF、MinNLCCreateF，且都带 SetBC/SetScale/SetCond 三步前置。 LM 比较特殊，它用 double 变量但走 MinLMCreateV 起手，且官方未提供 RequestTermination 方法（表中标注 does not exist），实盘里想中途撤优化得另想办法。 这张对照的价值在于：你按表里的「对象列表 + 调用方法列表」顺序抄进 EA，编译报错的概率会明显降低；外汇与贵金属策略参数寻优属高风险操作，回测拟合不代表实盘概率占优。
把重复跑分交给小布
这些诊断小布盯盘的 AIGC 已内置，打开对应品种页即可看到不同优化器在多维面上的收敛对比，你只需判断哪类约束更贴近自己的参数空间。
常见问题
BC 只处理盒式约束，约束激活策略更轻，大规模多活跃约束时可能比 BLEIC 更快；BLEIC 还能吃线性等式-不等式，通用性更高。
可以，小布内置了常见优化器的收敛与约束诊断视图，省去你手写脚本跑测试的重复劳动，你专注决策选型即可。
相对无约束方法鲁棒些，但非线性约束下初始点离可行域太远仍可能迭代失败，建议先做可行性探测再跑主优化。
LM 倾向用于残差平方和类最小二乘问题，参数不多且目标平滑时概率上收敛稳，但盒式边界需额外处理。
minbc 对应 BC 盒式约束优化，minnlc 对应基于预处理增广拉格朗日的 NLC，具体接口参数见文末函数表小节。