utility_i = alpha * quality_j - beta * price_j + gamma * (1/distance_j) + epsilon

graph TD
    A[初始化市场: N家医院, M位患者] --> B{每季度循环}
    B --> C[医院Agent规划: 调价/投资/退出决策]
    C --> D[患者Agent选择医院]
    D --> E[计算各医院利润、患者数量、质量变化]
    E --> F[保险公司更新保费]
    F --> G[记录成本指数 & 数据]
    G --> H{是否达到终止条件?}
    H -- 否 --> B
    H -- 是 --> I[输出模拟结果: 平均成本趋势, 不同收入群体影响]

def patient_utility(patient, hospital, hospitals):
    # 价格弹性系数随收入递减
    beta_base = 0.02 - 0.01 * (patient.income / max_income)
    distance_penalty = 0.1 * distance(patient.location, hospital.location)
    return (
        hospital.quality * (0.5 + 0.5 * patient.health_condition) -
        beta_base * hospital.price -
        distance_penalty +
        np.random.gumbel(0, 0.5)
    )

if profit_margin < target_margin * 0.8:
    if patient_volume > last_volume * 1.1:   # 需求有弹性
        price *= 1.05   # 温和涨价
    else:
        reduce_cost(by=0.02)
elif profit_margin > target_margin * 1.2:
    price *= 0.95   # 降价吸引患者

# requirements: mesa==2.1.4
import mesa
import numpy as np

class HospitalAgent(mesa.Agent):
    def __init__(self, unique_id, model, is_pe_owned):
        super().__init__(unique_id, model)
        self.is_pe_owned = is_pe_owned
        self.quality = np.random.uniform(0.5, 0.9)
        self.cost = np.random.uniform(80, 120)  # per visit
        self.price = self.cost * (1.2 if is_pe_owned else 1.1)
        self.margin_target = 0.20 if is_pe_owned else 0.0
        self.patients_last = 0

    def step(self):
        # 调整价格
        if self.is_pe_owned:
            current_margin = (self.price - self.cost) / self.price
            if current_margin < self.margin_target:
                # 尝试提价
                new_price = self.cost * (1 + self.margin_target)
                if new_price > self.price * 1.08:
                    new_price = self.price * 1.08
                self.price = new_price
            else:
                # 允许降价争取患者
                if np.random.random() < 0.3:
                    self.price *= 0.97

class PatientAgent(mesa.Agent):
    def __init__(self, unique_id, model, income, health):
        super().__init__(unique_id, model)
        self.income = income
        self.health = health
        self.chosen_hospital = None

    def step(self):
        # 选择效用最大的医院
        hospitals = self.model.get_agents_of_type(HospitalAgent)
        if not hospitals:
            return
        best_utility = -np.inf
        best_hospital = None
        for h in hospitals:
            # 简化：忽略距离
            utility = (h.quality * (0.5 + 0.5 * self.health) -
                       0.02 * (1 - self.income / 100000) * h.price +
                       np.random.gumbel(0, 0.5))
            if utility > best_utility:
                best_utility = utility
                best_hospital = h
        self.chosen_hospital = best_hospital

class HealthcareMarket(mesa.Model):
    def __init__(self, N_hospitals=10, N_patients=500, pe_ratio=0.3):
        super().__init__()
        self.N_hospitals = N_hospitals
        self.N_patients = N_patients
        # 创建医院Agent
        for i in range(N_hospitals):
            is_pe = i < int(N_hospitals * pe_ratio)  # 前几个为PE
            h = HospitalAgent(i, self, is_pe)
        # 创建患者Agent（收入分布正态，健康均匀）
        incomes = np.random.normal(60000, 20000, N_patients).clip(15000, 150000)
        healths = np.random.uniform(0.2, 0.8, N_patients)
        for j in range(N_patients):
            p = PatientAgent(N_hospitals + j, self, incomes[j], healths[j])
        # 数据收集
        self.datacollector = mesa.DataCollector(
            agent_reporters={"price": "price"}
        )

    def step(self):
        # 医院先行动
        self.agents_by_type[HospitalAgent].shuffle_do("step")
        # 患者再选择
        self.agents_by_type[PatientAgent].shuffle_do("step")
        # 收集数据
        self.datacollector.collect(self)

# 运行模拟
model = HealthcareMarket(pe_ratio=0.3)
for i in range(50):
    model.step()
    avg_price = np.mean([h.price for h in model.agents_by_type[HospitalAgent]])
    print(f"Step {i}: Average price = ${avg_price:.2f}")