全国 TOP 10 旅游城市节假日客流对比分析：K-Means 聚类 + 桑基图实战

🏙 全国 TOP 10 旅游城市名单:
1. 北京 (历史文化中心)
2. 上海 (国际金融中心)
3. 西安 (古都文明代表)
4. 成都 (休闲美食之都)
5. 杭州 (江南水乡典范)
6. 三亚 (热带滨海度假)
7. 重庆 (8D 魔幻山城)
8. 广州 (岭南文化枢纽)
9. 南京 (六朝古都)
10. 武汉 (九省通衢)
😤 传统对比分析的不足:
1⃣ 维度单一
   - 往往只比较总客流，忽略了人均消费、停留时长等质量指标
   - 缺乏对城市特质（如气候、IP）的深度考量
2⃣ 静态描述
   - 仅展示“是什么”，无法解释“为什么”
   - 缺乏对客流来源地和迁徙路径的动态追踪
3⃣ 缺乏分类指导
   - 不同类型的城市（如海滨 vs 山地）采用相同的运营策略
   - 无法实现精准的资源配置和政策制定

import pandas as pd
import numpy as np
from sklearn.cluster import KMeans
from sklearn.preprocessing import StandardScaler
import plotly.graph_objects as go
class CityDataLoader:
    """城市数据加载器"""
    def __init__(self):
        self.visitor_data = None
        self.consumption_data = None
        self.transport_data = None
    def load_visitor_data(self):
        """加载 10 个城市历史客流数据"""
        df = pd.read_csv('data/top10_cities_visitors_2023-2025.csv')
        print(f"客流数据量: {len(df)} 条")
        self.visitor_data = df
        return self
    def load_consumption_data(self):
        """加载消费水平数据"""
        df = pd.read_csv('data/cities_consumption_2023-2025.csv')
        # 字段：avg_hotel_price, avg_meal_cost, souvenir_spending
        print(f"消费数据量: {len(df)} 条")
        self.consumption_data = df
        return self
    def load_transport_data(self):
        """加载交通运力数据"""
        df = pd.read_csv('data/cities_transport_2023-2025.csv')
        # 字段：flight_arrivals, train_arrivals, highway_traffic
        print(f"交通数据量: {len(df)} 条")
        self.transport_data = df
        return self
    def merge_all_data(self):
        """合并所有数据源并构建指标体系"""
        df = self.visitor_data.merge(self.consumption_data, on=['city', 'date'], how='left')
        df = df.merge(self.transport_data, on=['city', 'date'], how='left')
        # 计算衍生指标
        df['revenue_per_visitor'] = df['total_revenue'] / df['total_visitors']
        df['avg_stay_duration'] = df['total_stay_days'] / df['total_visitors']
        return df
# 使用示例
loader = CityDataLoader
loader.load_visitor_data
loader.load_consumption_data
loader.load_transport_data
merged_data = loader.merge_all_data

from sklearn.metrics import silhouette_score
import matplotlib.pyplot as plt
def find_best_k(scaled_features, max_k=8):
    scores = []
    for k in range(2, max_k):
        kmeans = KMeans(n_clusters=k, random_state=42, n_init=10)
        labels = kmeans.fit_predict(scaled_features)
        score = silhouette_score(scaled_features, labels)
        scores.append(score)
        print(f"K={k}, Silhouette Score={score:.4f}")
    best_k = scores.index(max(scores)) + 2
    return best_k
# best_k = find_best_k(scaled_features)

class CityClusterAnalyzer:
    """城市聚类分析器"""
    def __init__(self, df):
        self.df = df.copy
        self.scaler = StandardScaler
        self.kmeans = None
    def prepare_features(self):
        """准备聚类特征"""
        features = self.df.groupby('city').agg({
            'total_visitors': 'mean',
            'revenue_per_visitor': 'mean',
            'avg_stay_duration': 'mean',
            'out_of_province_ratio': 'mean',
            'young_visitor_ratio': 'mean'
        }).reset_index
        feature_cols = ['total_visitors', 'revenue_per_visitor', 'avg_stay_duration',
                        'out_of_province_ratio', 'young_visitor_ratio']
        self.features = features[feature_cols]
        self.city_names = features['city']
        return self
    def fit_kmeans(self, n_clusters=3):
        """执行 K-Means 聚类"""
        scaled_features = self.scaler.fit_transform(self.features)
        self.kmeans = KMeans(n_clusters=n_clusters, random_state=42, n_init=10)
        self.labels = self.kmeans.fit_predict(scaled_features)
        self.features['cluster'] = self.labels
        self.features['city'] = self.city_names
        return self
    def get_cluster_profiles(self):
        """获取各类别的城市画像"""
        profiles = {}
        for cluster_id in range(3):
            cities_in_cluster = self.features[self.features['cluster'] == cluster_id]['city'].tolist
            profiles[f"Cluster {cluster_id}"] = cities_in_cluster
        return profiles
# 使用示例
analyzer = CityClusterAnalyzer(merged_data)
analyzer.prepare_features
analyzer.fit_kmeans(n_clusters=3)
profiles = analyzer.get_cluster_profiles
print(profiles)

def create_sankey_diagram(migration_data):
    """
    创建客流迁徙桑基图
    Args:
        migration_data: 包含 source, target, value 的数据框
    """
    # 获取所有唯一的节点名称
    all_nodes = list(set(migration_data['source'].tolist + migration_data['target'].tolist))
    node_indices = {node: i for i, node in enumerate(all_nodes)}
    fig = go.Figure(data=[go.Sankey(
        node = dict(
          pad = 15,
          thickness = 20,
          line = dict(color = "black", width = 0.5),
          label = all_nodes,
          color = "blue"
        ),
        link = dict(
          source = [node_indices[s] for s in migration_data['source']],
          target = [node_indices[t] for t in migration_data['target']],
          value = migration_data['value'],
          color = "rgba(0, 0, 255, 0.2)" # 半透明连线
      ))])
    fig.update_layout(title_text="五一假期全国主要城市客流迁徙路径", font_size=12, height=600)
    fig.show
# 使用示例：展示从主要省份流向 TOP 10 城市的客流
# create_sankey_diagram(migration_df)

from sklearn.metrics import silhouette_score
scores = []
for k in range(2, 8):
    kmeans = KMeans(n_clusters=k)
    labels = kmeans.fit_predict(scaled_features)
    score = silhouette_score(scaled_features, labels)
    scores.append(score)
best_k = scores.index(max(scores)) + 2  # 选择轮廓系数最高的 K

from sklearn.preprocessing import StandardScaler
scaler = StandardScaler
scaled_features = scaler.fit_transform(features)

go.Sankey(
    node=dict(pad=30, thickness=25),  # 增大间距
    link=dict(color='rgba(0,0,255,0.3)')  # 半透明连线
)

📋 城市运营建议:
□ 历史人文型城市：加强文物数字化展示，提升夜间游览体验
□ 自然风光型城市：完善极端天气下的应急预案，开发室内替代项目
□ 现代都市型城市：优化交通接驳网络，提升国际化服务水平