前言全球校园人工智能算法精英大赛”是江苏省人工智能学会举办的面向全球具有正式学籍的全日制高等院校及以上在校学生举办的算法竞赛。其中的算法巅峰赛属于产业命题赛道这是第3赛季这次优化题的主题是 “碳中和”。回顾第七届全球校园人工智能算法精英大赛-算法巅峰赛产业命题赛第3赛季优化题–碳中和对该文提到的 多策略模型提供具体的代码及解读。思路解析本算法保留经典贪心调度主框架并对核心逻辑进行策略抽象与解耦整体架构仅包含两类可插拔策略sort_strategy任务排序策略choice_strategy服务器挑选策略算法主流程依据指定的 sort_strategy 对任务列表进行全局排序按序遍历每一个任务遍历所有服务器筛选出分配后仍满足约束的候选服务器列表若候选列表非空则通过 choice_strategy 从中选定目标服务器将当前任务分配至选定的服务器执行。任务排序策略sort_strategy按收益值降序排序按收益 / 功耗比值降序排序按收益 / 热度比值降序排序按收益 /(功耗 热度) 比值降序排序按热度值升序排序服务器挑选策略choice_strategyFFFirst-Fit选择第一个满足约束的服务器BFBest-Fit按资源利用率来排序挑选服务器RFRandom-Fit从候选集中随机选择一台服务器带概率放弃的BF(Best-Fit)为最优挑选引入随机概率放弃多策略混合搜索机制将N 种任务排序策略与M 种服务器挑选策略进行全组合遍历形成 N×M 组调度方案。每组方案独立执行调度并计算得分最终选取总收益最高的方案作为最优解。设计特点全程以任务task主导分配结构清晰、易于实现与扩展策略完全解耦新增排序 / 挑选规则无需修改主框架组合搜索自动挖掘最优策略搭配大幅提升调度收益。代码多策略组合(代码由 Deepseek 生成)#includebits/stdc.husingnamespacestd;structTask{doublegain,power,heat;};structServer{doublepower_limit,heat_limit;};// 你原题的热约束完全不动classHeatChecker{private:constvectorServerservers;doublek;intm;public:HeatChecker(constvectorServerservers,doublek):servers(servers),k(k),m(servers.size()){}boolcanPlace(constvectordoublepowers,constvectordoubleheats,intpos,constTasktask)const{if(task.powerpowers[pos]servers[pos].power_limit1e-9)returnfalse;doubleself_heatheats[pos]task.heat;doublelh(pos0)?heats[pos-1]*k:0.0;doublerh(posm-1)?heats[pos1]*k:0.0;if(self_heatlhrhservers[pos].heat_limit1e-9)returnfalse;if(pos0){doubleleft_selfheats[pos-1];doubleleft_influenceleft_self(heats[pos]task.heat)*k;if(pos-10)left_influenceheats[pos-2]*k;if(left_influenceservers[pos-1].heat_limit1e-9)returnfalse;}if(posm-1){doubleright_selfheats[pos1];doubleright_influenceright_self(heats[pos]task.heat)*k;if(pos1m-1)right_influenceheats[pos2]*k;if(right_influenceservers[pos1].heat_limit1e-9)returnfalse;}returntrue;}};// 策略模式排序策略 classSortStrategy{public:virtualvectorintsortTasks(constvectorTasktasks)const0;virtual~SortStrategy()default;};classSortByGainDesc:publicSortStrategy{public:vectorintsortTasks(constvectorTasktasks)constoverride{vectorintorder(tasks.size());iota(order.begin(),order.end(),0);sort(order.begin(),order.end(),[](inta,intb){returntasks[a].gaintasks[b].gain;});returnorder;}};classSortByGainPowerRatio:publicSortStrategy{public:vectorintsortTasks(constvectorTasktasks)constoverride{vectorintorder(tasks.size());iota(order.begin(),order.end(),0);sort(order.begin(),order.end(),[](inta,intb){returntasks[a].gain/tasks[a].powertasks[b].gain/tasks[b].power;});returnorder;}};classSortByGainHeatRatio:publicSortStrategy{public:vectorintsortTasks(constvectorTasktasks)constoverride{vectorintorder(tasks.size());iota(order.begin(),order.end(),0);sort(order.begin(),order.end(),[](inta,intb){returntasks[a].gain/tasks[a].heattasks[b].gain/tasks[b].heat;});returnorder;}};classSortByGainTotalRatio:publicSortStrategy{public:vectorintsortTasks(constvectorTasktasks)constoverride{vectorintorder(tasks.size());iota(order.begin(),order.end(),0);sort(order.begin(),order.end(),[](inta,intb){doubleratasks[a].gain/(tasks[a].powertasks[a].heat);doublerbtasks[b].gain/(tasks[b].powertasks[b].heat);returnrarb;});returnorder;}};classSortByHeatAsc:publicSortStrategy{public:vectorintsortTasks(constvectorTasktasks)constoverride{vectorintorder(tasks.size());iota(order.begin(),order.end(),0);sort(order.begin(),order.end(),[](inta,intb){returntasks[a].heattasks[b].heat;});returnorder;}};// 策略模式挑选策略 classChoiceStrategy{public:virtualintselect(constvectorintcandidates,constvectordoublecur_p,constvectordoublecur_h,constvectorServerservers,doublek,constTasktask)const0;virtual~ChoiceStrategy()default;};// FF 首次适应classFirstFit:publicChoiceStrategy{public:intselect(constvectorintcandidates,constvectordouble,constvectordouble,constvectorServer,double,constTask)constoverride{returncandidates.empty()?-1:candidates[0];}};// 【你专属的最优适应】完全用你的公式 classBestFit:publicChoiceStrategy{public:intselect(constvectorintcandidates,constvectordoublepowers,constvectordoubleheats,constvectorServerservers,doublek,constTasktask)constoverride{if(candidates.empty())return-1;intbest_pos-1;doublemin_score1e18;for(intpos:candidates){doublepower_ratio(powers[pos]task.power)/servers[pos].power_limit;doubleheat_ratio(heats[pos]task.heat)/servers[pos].heat_limit;doubleheat_spreadheats[pos]task.heat;if(pos0)heat_spreadheats[pos-1]*k;if(pos(int)servers.size()-1)heat_spreadheats[pos1]*k;doubleheat_ratio_with_spreadheat_spread/servers[pos].heat_limit;doublescoremax(power_ratio,heat_ratio_with_spread);doubleremainingmin(servers[pos].power_limit-powers[pos]-task.power,servers[pos].heat_limit-heats[pos]-task.heat);score-remaining/(servers[pos].power_limitservers[pos].heat_limit)*0.1;// 分数越小越好if(scoremin_score){min_scorescore;best_pospos;}}returnbest_pos;}};// 调度器 pairdouble,vectorintschedule(constvectorTasktasks,constvectorServerservers,doublek,constSortStrategysort_strategy,constChoiceStrategychoice_strategy){intntasks.size();intmservers.size();vectorintordersort_strategy.sortTasks(tasks);HeatCheckerchecker(servers,k);vectorintans(n,-1);vectordoublecur_p(m,0),cur_h(m,0);doubletotal_gain0;for(intid:order){constTaskttasks[id];vectorintcandidates;for(intj0;jm;j){if(checker.canPlace(cur_p,cur_h,j,t)){candidates.push_back(j);}}intschoice_strategy.select(candidates,cur_p,cur_h,servers,k,t);if(s!-1){ans[id]s;cur_p[s]t.power;cur_h[s]t.heat;total_gaint.gain;}}return{total_gain,ans};}// 主函数策略组合遍历 最优选择 intmain(){ios::sync_with_stdio(false);cin.tie(0);intn,m;doublek;cinnmk;vectorTasktasks(n);for(inti0;in;i){cintasks[i].gaintasks[i].powertasks[i].heat;}vectorServerservers(m);for(inti0;im;i){cinservers[i].power_limitservers[i].heat_limit;}vectorunique_ptrSortStrategysorts;sorts.emplace_back(newSortByGainDesc());sorts.emplace_back(newSortByGainPowerRatio());sorts.emplace_back(newSortByGainHeatRatio());sorts.emplace_back(newSortByGainTotalRatio());sorts.emplace_back(newSortByHeatAsc());vectorunique_ptrChoiceStrategychoices;choices.emplace_back(newFirstFit());choices.emplace_back(newBestFit());doublebest_gain-1;vectorintbest_ans(n,-1);for(autos:sorts){for(autoc:choices){auto[gain,ans]schedule(tasks,servers,k,*s,*c);if(gainbest_gain){best_gaingain;best_ansans;}}}for(inti0;in;i){coutbest_ans[i]1 \n[in-1];}return0;}效果评估在基准测试集中这个分值接近赛时第一的分数我猜测他可能用了类似解法。这个算法中贡献最大的是按 收益/热度 比值 降序排序采用 最优挑选策略选择服务器写在最后