贪心算法应用:装箱问题(FFD问题)详解
1. 装箱问题概述
装箱问题(Bin Packing Problem)是计算机科学和运筹学中的一个经典组合优化问题。问题的描述如下:
给定一组物品,每个物品有一定的体积,以及若干容量相同的箱子,目标是用最少数量的箱子装下所有物品。
问题形式化描述
- 输入:
- n个物品,每个物品有一个大小wᵢ,其中0 < wᵢ ≤ C(C为箱子容量)
- 无限数量的箱子,每个箱子容量为C
- 输出:
- 将n个物品分配到尽可能少的箱子中,且每个箱子中物品大小之和不超过C
2. 贪心算法简介
贪心算法(Greedy Algorithm)是一种在每一步选择中都采取在当前状态下最好或最优(即最有利)的选择,从而希望导致结果是全局最好或最优的算法。
对于装箱问题,常见的贪心算法策略有:
- 首次适应算法(First Fit, FF):将每个物品放入第一个能容纳它的箱子
- 最佳适应算法(Best Fit, BF):将每个物品放入能容纳它的最满的箱子
- 首次适应递减算法(First Fit Decreasing, FFD):先将物品按大小降序排序,然后使用首次适应算法
- 最佳适应递减算法(Best Fit Decreasing, BFD):先将物品按大小降序排序,然后使用最佳适应算法
本文将重点介绍**首次适应递减算法(FFD)**及其Java实现。
3. 首次适应递减算法(FFD)详解
3.1 算法思想
FFD算法是解决装箱问题最常用的启发式算法之一,其基本思想是:
- 先将所有物品按体积从大到小排序
- 然后依次处理每个物品,将其放入第一个能容纳它的箱子
- 如果没有合适的箱子,则开启一个新箱子
3.2 算法步骤
- 输入物品列表和箱子容量C
- 将物品按体积从大到小排序
- 初始化空的箱子列表
- 对于每个物品:
a. 遍历已有箱子,找到第一个能容纳该物品的箱子
b. 如果找到,将物品放入该箱子
c. 如果没有找到,创建一个新箱子并将物品放入 - 返回使用的箱子列表
3.3 算法复杂度分析
- 排序阶段:O(n log n),取决于排序算法
- 装箱阶段:最坏情况下为O(n²),因为对于每个物品可能需要遍历所有箱子
3.4 算法性能
FFD算法有以下性能保证:
- 对于任何输入,FFD使用的箱子数不超过(11/9)*OPT + 1,其中OPT是最优解
- 对于大多数实际案例,FFD的表现非常接近最优解
4. Java实现FFD算法
4.1 基本实现
import java.util.ArrayList;
import java.util.Collections;
import java.util.List;public class BinPackingFFD {public static void main(String[] args) {// 示例物品大小List<Integer> items = List.of(4, 8, 5, 1, 2, 3, 6, 7, 9, 4);int binCapacity = 10;List<List<Integer>> bins = firstFitDecreasing(items, binCapacity);System.out.println("使用的箱子数量: " + bins.size());for (int i = 0; i < bins.size(); i++) {System.out.println("箱子 " + (i+1) + ": " + bins.get(i) + " (总大小: " + bins.get(i).stream().mapToInt(Integer::intValue).sum() + ")");}}public static List<List<Integer>> firstFitDecreasing(List<Integer> items, int binCapacity) {// 复制物品列表以避免修改原始数据List<Integer> sortedItems = new ArrayList<>(items);// 按降序排序sortedItems.sort(Collections.reverseOrder());List<List<Integer>> bins = new ArrayList<>();for (int item : sortedItems) {boolean placed = false;// 尝试将物品放入已有箱子for (List<Integer> bin : bins) {int currentBinWeight = bin.stream().mapToInt(Integer::intValue).sum();if (currentBinWeight + item <= binCapacity) {bin.add(item);placed = true;break;}}// 如果没有合适的箱子,创建新箱子if (!placed) {List<Integer> newBin = new ArrayList<>();newBin.add(item);bins.add(newBin);}}return bins;}
}
4.2 优化实现
为了提高效率,我们可以预先计算并存储每个箱子的剩余容量:
import java.util.ArrayList;
import java.util.Collections;
import java.util.List;public class BinPackingFFDOptimized {public static void main(String[] args) {List<Integer> items = List.of(4, 8, 5, 1, 2, 3, 6, 7, 9, 4);int binCapacity = 10;List<Bin> bins = firstFitDecreasingOptimized(items, binCapacity);System.out.println("使用的箱子数量: " + bins.size());for (int i = 0; i < bins.size(); i++) {System.out.println("箱子 " + (i+1) + ": " + bins.get(i).items + " (总大小: " + bins.get(i).currentWeight + ")");}}static class Bin {List<Integer> items = new ArrayList<>();int currentWeight = 0;int capacity;Bin(int capacity) {this.capacity = capacity;}boolean canAdd(int item) {return currentWeight + item <= capacity;}void addItem(int item) {items.add(item);currentWeight += item;}}public static List<Bin> firstFitDecreasingOptimized(List<Integer> items, int binCapacity) {List<Integer> sortedItems = new ArrayList<>(items);sortedItems.sort(Collections.reverseOrder());List<Bin> bins = new ArrayList<>();for (int item : sortedItems) {boolean placed = false;for (Bin bin : bins) {if (bin.canAdd(item)) {bin.addItem(item);placed = true;break;}}if (!placed) {Bin newBin = new Bin(binCapacity);newBin.addItem(item);bins.add(newBin);}}return bins;}
}
4.3 进一步优化:使用优先队列
我们可以使用优先队列来更高效地找到合适的箱子:
import java.util.ArrayList;
import java.util.Collections;
import java.util.List;
import java.util.PriorityQueue;public class BinPackingFFDWithPQ {public static void main(String[] args) {List<Integer> items = List.of(4, 8, 5, 1, 2, 3, 6, 7, 9, 4);int binCapacity = 10;List<Bin> bins = firstFitDecreasingWithPQ(items, binCapacity);System.out.println("使用的箱子数量: " + bins.size());for (int i = 0; i < bins.size(); i++) {System.out.println("箱子 " + (i+1) + ": " + bins.get(i).items + " (总大小: " + bins.get(i).currentWeight + ")");}}static class Bin implements Comparable<Bin> {List<Integer> items = new ArrayList<>();int currentWeight = 0;int capacity;Bin(int capacity) {this.capacity = capacity;}boolean canAdd(int item) {return currentWeight + item <= capacity;}void addItem(int item) {items.add(item);currentWeight += item;}// 按照剩余容量升序排列,这样我们可以优先尝试剩余容量多的箱子@Overridepublic int compareTo(Bin other) {return Integer.compare(other.capacity - other.currentWeight, this.capacity - this.currentWeight);}}public static List<Bin> firstFitDecreasingWithPQ(List<Integer> items, int binCapacity) {List<Integer> sortedItems = new ArrayList<>(items);sortedItems.sort(Collections.reverseOrder());List<Bin> bins = new ArrayList<>();PriorityQueue<Bin> pq = new PriorityQueue<>();for (int item : sortedItems) {Bin bin = pq.peek();if (bin != null && bin.canAdd(item)) {bin = pq.poll();bin.addItem(item);pq.offer(bin);} else {Bin newBin = new Bin(binCapacity);newBin.addItem(item);bins.add(newBin);pq.offer(newBin);}}return bins;}
}
5. 算法测试与验证
5.1 测试用例设计
为了验证我们的实现是否正确,我们可以设计以下测试用例:
-
简单测试:少量物品,容易验证
- 输入:[2, 3, 4, 5], 容量=7
- 预期:2个箱子 [5,2]和[4,3]
-
边界测试:
- 所有物品大小相同
- 单个物品正好装满一个箱子
- 单个物品超过箱子容量(应抛出异常)
-
随机测试:
- 生成随机物品列表进行测试
-
已知最优解测试:
- 使用已知最优解的小规模问题
5.2 测试代码实现
import org.junit.Test;
import static org.junit.Assert.*;
import java.util.List;public class BinPackingFFDTest {@Testpublic void testSimpleCase() {List<Integer> items = List.of(2, 3, 4, 5);int binCapacity = 7;List<List<Integer>> bins = BinPackingFFD.firstFitDecreasing(items, binCapacity);assertEquals(2, bins.size());assertTrue(bins.get(0).containsAll(List.of(5, 2)) || bins.get(1).containsAll(List.of(5, 2)));assertTrue(bins.get(0).containsAll(List.of(4, 3)) || bins.get(1).containsAll(List.of(4, 3)));}@Testpublic void testPerfectFit() {List<Integer> items = List.of(5, 5, 5, 5);int binCapacity = 10;List<List<Integer>> bins = BinPackingFFD.firstFitDecreasing(items, binCapacity);assertEquals(2, bins.size());for (List<Integer> bin : bins) {assertEquals(10, bin.stream().mapToInt(Integer::intValue).sum());}}@Testpublic void testSingleItem() {List<Integer> items = List.of(7);int binCapacity = 10;List<List<Integer>> bins = BinPackingFFD.firstFitDecreasing(items, binCapacity);assertEquals(1, bins.size());assertEquals(7, bins.get(0).stream().mapToInt(Integer::intValue).sum());}@Test(expected = IllegalArgumentException.class)public void testItemTooLarge() {List<Integer> items = List.of(11);int binCapacity = 10;BinPackingFFD.firstFitDecreasing(items, binCapacity);}@Testpublic void testEmptyInput() {List<Integer> items = List.of();int binCapacity = 10;List<List<Integer>> bins = BinPackingFFD.firstFitDecreasing(items, binCapacity);assertTrue(bins.isEmpty());}
}
6. 性能分析与优化
6.1 时间复杂度分析
- 排序阶段:O(n log n)
- 装箱阶段:
- 基本实现:O(n²) - 对于每个物品,最坏情况下需要检查所有箱子
- 优先队列优化:O(n log n) - 每次插入和提取操作都是O(log n)
6.2 空间复杂度分析
- O(n) - 需要存储所有物品和箱子信息
6.3 实际性能测试
我们可以编写性能测试代码来比较不同实现的性能:
import java.util.ArrayList;
import java.util.Collections;
import java.util.List;
import java.util.Random;public class BinPackingPerformanceTest {public static void main(String[] args) {int numItems = 10000;int binCapacity = 100;List<Integer> items = generateRandomItems(numItems, binCapacity);// 预热firstFitDecreasing(new ArrayList<>(items), binCapacity);firstFitDecreasingOptimized(new ArrayList<>(items), binCapacity);firstFitDecreasingWithPQ(new ArrayList<>(items), binCapacity);// 测试基本实现long start = System.currentTimeMillis();List<List<Integer>> bins1 = firstFitDecreasing(new ArrayList<>(items), binCapacity);long end = System.currentTimeMillis();System.out.println("基本实现: " + (end - start) + "ms, 箱子数: " + bins1.size());// 测试优化实现start = System.currentTimeMillis();List<BinPackingFFDOptimized.Bin> bins2 = firstFitDecreasingOptimized(new ArrayList<>(items), binCapacity);end = System.currentTimeMillis();System.out.println("优化实现: " + (end - start) + "ms, 箱子数: " + bins2.size());// 测试优先队列实现start = System.currentTimeMillis();List<BinPackingFFDWithPQ.Bin> bins3 = firstFitDecreasingWithPQ(new ArrayList<>(items), binCapacity);end = System.currentTimeMillis();System.out.println("优先队列实现: " + (end - start) + "ms, 箱子数: " + bins3.size());}private static List<Integer> generateRandomItems(int numItems, int maxSize) {List<Integer> items = new ArrayList<>();Random random = new Random();for (int i = 0; i < numItems; i++) {items.add(random.nextInt(maxSize) + 1); // 1到maxSize}return items;}// 这里需要包含前面三个实现的方法...
}
7. 应用场景与扩展
7.1 实际应用场景
装箱问题在现实世界中有许多应用:
- 物流与运输:将货物装入集装箱或卡车
- 资源分配:云计算中的虚拟机分配
- 存储管理:文件存储到磁盘或内存中
- 生产计划:任务分配到机器上
- 广告投放:将广告分配到固定时长的广告位
7.2 变种与扩展
- 多维装箱问题:物品有多个维度(长、宽、高)
- 可变大小箱子:箱子大小可以不同
- 成本最小化:不同箱子有不同的成本
- 在线装箱问题:物品按顺序到达,必须立即分配
- 带冲突的装箱问题:某些物品不能放在同一个箱子中
7.3 其他算法比较
虽然FFD是一个很好的启发式算法,但还有其他算法可以解决装箱问题:
-
精确算法:
- 分支限界法
- 动态规划(适用于小规模问题)
-
近似算法:
- Next Fit (NF)
- Worst Fit (WF)
- Almost Worst Fit (AWF)
-
元启发式算法(适用于大规模问题):
- 遗传算法
- 模拟退火
- 禁忌搜索
8. 完整Java实现示例
以下是结合了所有优化和功能的完整实现:
import java.util.ArrayList;
import java.util.Collections;
import java.util.List;
import java.util.PriorityQueue;public class AdvancedBinPackingFFD {public static void main(String[] args) {// 示例使用List<Integer> items = generateRandomItems(20, 10);int binCapacity = 10;System.out.println("物品列表: " + items);BinPackingResult result = packItems(items, binCapacity);System.out.println("使用的箱子数量: " + result.getBinCount());System.out.println("平均填充率: " + String.format("%.2f", result.getAverageFillRate() * 100) + "%");System.out.println("详细装箱情况:");result.printBins();}/*** 装箱结果类*/public static class BinPackingResult {private final List<Bin> bins;private final int binCapacity;public BinPackingResult(List<Bin> bins, int binCapacity) {this.bins = bins;this.binCapacity = binCapacity;}public int getBinCount() {return bins.size();}public double getAverageFillRate() {return bins.stream().mapToDouble(bin -> (double)bin.getCurrentWeight() / binCapacity).average().orElse(0);}public void printBins() {for (int i = 0; i < bins.size(); i++) {Bin bin = bins.get(i);System.out.printf("箱子 %2d: %s (总大小: %2d, 填充率: %5.2f%%)%n",i + 1, bin.getItems(), bin.getCurrentWeight(),(double)bin.getCurrentWeight() / binCapacity * 100);}}public List<Bin> getBins() {return Collections.unmodifiableList(bins);}}/*** 箱子类*/public static class Bin implements Comparable<Bin> {private final List<Integer> items = new ArrayList<>();private int currentWeight = 0;private final int capacity;public Bin(int capacity) {this.capacity = capacity;}public boolean canAdd(int item) {if (item > capacity) {throw new IllegalArgumentException("物品大小超过箱子容量");}return currentWeight + item <= capacity;}public void addItem(int item) {if (!canAdd(item)) {throw new IllegalStateException("无法将物品添加到箱子中");}items.add(item);currentWeight += item;}public List<Integer> getItems() {return Collections.unmodifiableList(items);}public int getCurrentWeight() {return currentWeight;}public int getRemainingCapacity() {return capacity - currentWeight;}@Overridepublic int compareTo(Bin other) {// 按剩余容量降序排列return Integer.compare(other.getRemainingCapacity(), this.getRemainingCapacity());}}/*** 装箱方法*/public static BinPackingResult packItems(List<Integer> items, int binCapacity) {// 验证输入if (binCapacity <= 0) {throw new IllegalArgumentException("箱子容量必须为正数");}for (int item : items) {if (item <= 0) {throw new IllegalArgumentException("物品大小必须为正数");}if (item > binCapacity) {throw new IllegalArgumentException("存在物品大小超过箱子容量");}}// 复制物品列表以避免修改原始数据List<Integer> sortedItems = new ArrayList<>(items);// 按降序排序sortedItems.sort(Collections.reverseOrder());List<Bin> bins = new ArrayList<>();PriorityQueue<Bin> binQueue = new PriorityQueue<>();for (int item : sortedItems) {Bin bin = binQueue.peek();if (bin != null && bin.canAdd(item)) {bin = binQueue.poll();bin.addItem(item);binQueue.offer(bin);} else {Bin newBin = new Bin(binCapacity);newBin.addItem(item);bins.add(newBin);binQueue.offer(newBin);}}return new BinPackingResult(bins, binCapacity);}/*** 生成随机物品列表*/public static List<Integer> generateRandomItems(int count, int maxSize) {List<Integer> items = new ArrayList<>();java.util.Random random = new java.util.Random();for (int i = 0; i < count; i++) {items.add(random.nextInt(maxSize) + 1); // 1到maxSize}return items;}
}
9. 总结
首次适应递减算法(FFD)是解决装箱问题的一种高效启发式算法,通过先将物品按大小降序排序,然后使用首次适应策略,能够在大多数情况下得到接近最优的解。本文详细介绍了:
- 装箱问题的定义和贪心算法的基本概念
- FFD算法的详细思想和实现步骤
- 多种Java实现方式,包括基本实现、优化实现和使用优先队列的实现
- 测试用例设计和性能分析方法
- 实际应用场景和算法扩展
FFD算法的时间复杂度主要取决于排序阶段(O(n log n))和装箱阶段(O(n²)或优化后的O(n log n)),在实际应用中表现良好。对于需要更高精度的场景,可以考虑结合其他优化算法或精确算法。