""" 模拟真实BTC/ETH走势的测试数据生成器 基于BTC 2017-2025年的实际周期结构: - 2017 Q3-Q4: 牛市 $4K → $20K - 2018: 熊市 $20K → $3.2K - 2019: 复苏 $3.2K → $14K → 回落$7K - 2020 Q1-Q3: COVID崩盘+复苏 $7K→$3.8K→$12K - 2020 Q4-2021 Q2: 大牛 $12K → $65K - 2021 Q3: 回调 $65K → $30K - 2021 Q4: 二次高点 $30K → $69K - 2022: 大熊 $69K → $15.5K - 2023: 慢牛 $15.5K → $44K - 2024: ETF牛市 $44K → $73K → 回调 → $100K - 2025 Q1: $100K → $80K → $90K 每个阶段的关键: 跌幅、盒子（横盘）、放量突破 """ import numpy as np import pandas as pd from typing import List, Tuple def _generate_phase( start_price: float, end_price: float, n_hours: int, volatility: float = 0.003, box_zones: List[Tuple[float, float, int]] = None, ) -> np.ndarray: """ 生成单个阶段的小时价格 box_zones: [(start_frac, end_frac, box_duration_hours)] 在阶段的start_frac~end_frac位置插入横盘盒子 """ np.random.seed(abs(hash(f"{start_price}{end_price}{n_hours}")) % 2**31) prices = np.zeros(n_hours) prices[0] = start_price # 基础漂移 log_return = np.log(end_price / start_price) drift_per_hour = log_return / n_hours for i in range(1, n_hours): frac = i / n_hours local_drift = drift_per_hour # 在盒子区域压缩drift到接近0 if box_zones: for bstart, bend, _ in box_zones: if bstart <= frac <= bend: local_drift *= 0.05 # 横盘时几乎不动 noise = np.random.normal(0, volatility) prices[i] = prices[i-1] * np.exp(local_drift + noise) # 确保终点正确 prices = prices * (end_price / prices[-1]) ** (np.linspace(0, 1, n_hours) ** 3) return prices def _add_volume_pattern(n: int, base_vol: float = 500, spike_positions: List[float] = None) -> np.ndarray: """生成带有放量特征的成交量""" np.random.seed(42 + n) vol = np.random.lognormal(np.log(base_vol), 0.3, n) # 在指定位置加放量 if spike_positions: for pos in spike_positions: idx = int(pos * n) for j in range(max(0, idx-5), min(n, idx+10)): vol[j] *= np.random.uniform(2.5, 5.0) return vol def generate_btc_history(start_date: str = '2017-08-01', end_date: str = '2026-03-31') -> pd.DataFrame: """ 生成模拟BTC 1H K线数据 基于真实价格结构，包含: - 大跌后的盒子（横盘区域） - 突破时的放量 - 真实的波动率变化 """ phases = [ # (end_price, hours, vol, box_zones, vol_spikes) # 2017 Aug-Dec: 牛市 (4000, 5000, 200, 0.006, None, [0.3, 0.6, 0.9]), # $4K→$20K # 这里分细一点 ] # 更精确的分段定义 segments = [ # (start_price, end_price, start_date, end_date, volatility, has_box_and_breakout) # 2017 (4000, 6500, '2017-08-01', '2017-10-01', 0.004, False), (6500, 7800, '2017-10-01', '2017-11-01', 0.005, False), (7800, 19500, '2017-11-01', '2017-12-17', 0.008, False), (19500, 13500, '2017-12-17', '2018-01-01', 0.010, False), # 2018 大熊 (13500, 11500, '2018-01-01', '2018-02-06', 0.008, False), (11500, 8500, '2018-02-06', '2018-03-30', 0.006, True), # 盒子 (8500, 6200, '2018-03-30', '2018-06-30', 0.005, True), # 盒子 (6200, 3200, '2018-06-30', '2018-12-15', 0.004, True), # 盒子 # 2019 复苏 (3200, 3500, '2018-12-15', '2019-03-01', 0.003, True), # 底部盒子! (3500, 5500, '2019-03-01', '2019-04-02', 0.005, False), # 突破 (5500, 8500, '2019-04-02', '2019-05-15', 0.006, False), (8500, 13800, '2019-05-15', '2019-06-26', 0.007, False), (13800, 9500, '2019-06-26', '2019-09-30', 0.005, True), # 回调+盒子 (9500, 7200, '2019-09-30', '2019-12-31', 0.004, True), # 盒子 # 2020 (7200, 9200, '2020-01-01', '2020-02-14', 0.004, False), (9200, 3800, '2020-02-14', '2020-03-16', 0.015, False), # COVID崩盘 (3800, 5000, '2020-03-16', '2020-04-15', 0.006, True), # 底部盒子! (5000, 9500, '2020-04-15', '2020-07-20', 0.004, True), # 盒子→突破 (9500, 10500, '2020-07-20', '2020-10-01', 0.003, True), # 盒子! (10500, 19800, '2020-10-01', '2020-12-01', 0.005, False), # 突破+大涨 (19800, 29000, '2020-12-01', '2021-01-08', 0.007, False), # 2021 牛市 (29000, 42000, '2021-01-08', '2021-02-21', 0.006, False), (42000, 58000, '2021-02-21', '2021-03-13', 0.008, False), (58000, 64900, '2021-03-13', '2021-04-14', 0.006, False), # ATH1 (64900, 30000, '2021-04-14', '2021-06-22', 0.008, True), # 大回调 (30000, 35000, '2021-06-22', '2021-07-20', 0.005, True), # 盒子! (35000, 52000, '2021-07-20', '2021-09-07', 0.005, False), # 突破 (52000, 41000, '2021-09-07', '2021-09-29', 0.007, False), (41000, 69000, '2021-09-29', '2021-11-10', 0.006, False), # ATH2 (69000, 46000, '2021-11-10', '2021-12-31', 0.007, False), # 2022 大熊 (46000, 38000, '2022-01-01', '2022-02-28', 0.005, True), # 盒子 (38000, 30000, '2022-02-28', '2022-05-09', 0.004, True), # 盒子 (30000, 18000, '2022-05-09', '2022-06-18', 0.010, False), # LUNA崩盘 (18000, 20000, '2022-06-18', '2022-08-15', 0.003, True), # 底部盒子 (20000, 24500, '2022-08-15', '2022-08-25', 0.005, False), # 小反弹 (24500, 19500, '2022-08-25', '2022-09-20', 0.005, False), (19500, 20500, '2022-09-20', '2022-11-05', 0.003, True), # 盒子 (20500, 15500, '2022-11-05', '2022-11-21', 0.012, False), # FTX崩盘 (15500, 16500, '2022-11-21', '2022-12-31', 0.003, True), # 底部盒子! # 2023 慢牛 (16500, 21000, '2023-01-01', '2023-01-20', 0.005, False), # 突破 (21000, 23500, '2023-01-20', '2023-02-20', 0.004, False), (23500, 28500, '2023-02-20', '2023-03-20', 0.005, False), (28500, 25000, '2023-03-20', '2023-06-15', 0.003, True), # 盒子 (25000, 31500, '2023-06-15', '2023-07-13', 0.005, False), # 突破 (31500, 26000, '2023-07-13', '2023-09-11', 0.003, True), # 盒子 (26000, 35000, '2023-09-11', '2023-10-24', 0.005, False), # 突破 (35000, 38000, '2023-10-24', '2023-11-15', 0.004, False), (38000, 44000, '2023-11-15', '2023-12-31', 0.004, False), # 2024 ETF牛市 (44000, 49000, '2024-01-01', '2024-01-11', 0.005, False), # ETF批准 (49000, 52000, '2024-01-11', '2024-02-07', 0.004, False), (52000, 69000, '2024-02-07', '2024-03-14', 0.006, False), # 新ATH (69000, 73800, '2024-03-14', '2024-03-15', 0.008, False), # ATH (73800, 57000, '2024-03-15', '2024-05-01', 0.005, True), # 回调 (57000, 60000, '2024-05-01', '2024-07-15', 0.003, True), # 盒子 (60000, 72000, '2024-07-15', '2024-10-29', 0.004, False), # 突破 (72000, 100000, '2024-10-29', '2024-12-05', 0.006, False), # 大选行情 (100000, 92000, '2024-12-05', '2024-12-31', 0.005, False), # 2025 (92000, 106000, '2025-01-01', '2025-01-20', 0.005, False), (106000, 78000, '2025-01-20', '2025-02-28', 0.006, False), # 回调 (78000, 84000, '2025-02-28', '2025-03-31', 0.004, True), # 盒子 ] all_prices = [] all_volumes = [] all_timestamps = [] for seg in segments: s_price, e_price, s_date, e_date, vol, has_box = seg s_ts = pd.Timestamp(s_date, tz='UTC') e_ts = pd.Timestamp(e_date, tz='UTC') n_hours = int((e_ts - s_ts).total_seconds() / 3600) if n_hours <= 0: continue # 生成价格 box_zones = [(0.2, 0.5, n_hours // 3)] if has_box else None prices = _generate_phase(s_price, e_price, n_hours, vol, box_zones) # 生成成交量（突破点放量） spike_pos = [0.5] if has_box else [0.3, 0.7] volumes = _add_volume_pattern(n_hours, 300 + abs(e_price - s_price) * 0.01, spike_pos) timestamps = pd.date_range(s_date, periods=n_hours, freq='1h', tz='UTC') all_prices.extend(prices) all_volumes.extend(volumes) all_timestamps.extend(timestamps) prices = np.array(all_prices) volumes = np.array(all_volumes) n = len(prices) np.random.seed(12345) df = pd.DataFrame({ 'timestamp': all_timestamps[:n], 'open': prices * (1 + np.random.normal(0, 0.001, n)), 'high': prices * (1 + np.abs(np.random.normal(0.003, 0.002, n))), 'low': prices * (1 - np.abs(np.random.normal(0.003, 0.002, n))), 'close': prices, 'volume': volumes, }) # 去重保留最后一个 df = df.drop_duplicates(subset='timestamp', keep='last').sort_values('timestamp').reset_index(drop=True) return df def generate_eth_history(btc_df: pd.DataFrame) -> pd.DataFrame: """ 基于BTC走势生成ETH走势 ETH和BTC高度相关，但波动更大（beta ≈ 1.3-1.5） """ np.random.seed(99999) btc_prices = btc_df['close'].values n = len(btc_prices) # ETH/BTC ratio 历史走势 # 2017: ~0.05 2018: 0.03 2019: 0.02 2020: 0.03 2021: 0.04-0.08 # 2022: 0.06 2023: 0.06 2024: 0.04-0.05 ratio_anchors = { 0.00: 0.075, # 2017 Aug: ETH~$300 / BTC~$4000 0.10: 0.060, 0.20: 0.035, # 2018 bear 0.30: 0.020, # 2019 low ratio 0.40: 0.030, 0.50: 0.040, # 2020 DeFi summer 0.60: 0.070, # 2021 bull 0.70: 0.080, # 2021 peak 0.80: 0.065, # 2022-2023 0.90: 0.045, # 2024 1.00: 0.035, # 2025 } fracs = np.array(sorted(ratio_anchors.keys())) ratios = np.array([ratio_anchors[f] for f in fracs]) x = np.linspace(0, 1, n) interp_ratio = np.interp(x, fracs, ratios) # Add noise to ratio ratio_noise = np.random.normal(0, 0.002, n) ratio_smooth = interp_ratio + np.cumsum(ratio_noise) * 0.0001 eth_prices = btc_prices * ratio_smooth # Make ETH more volatile (higher beta) btc_returns = np.diff(np.log(btc_prices)) eth_excess = btc_returns * np.random.uniform(0.2, 0.6, len(btc_returns)) eth_prices_adj = np.zeros(n) eth_prices_adj[0] = eth_prices[0] for i in range(1, n): base_return = np.log(eth_prices[i] / eth_prices[i-1]) eth_prices_adj[i] = eth_prices_adj[i-1] * np.exp(base_return + np.random.normal(0, 0.002)) volumes = _add_volume_pattern(n, 500, [0.3, 0.5, 0.7]) np.random.seed(54321) df = pd.DataFrame({ 'timestamp': btc_df['timestamp'].values[:n], 'open': eth_prices_adj * (1 + np.random.normal(0, 0.001, n)), 'high': eth_prices_adj * (1 + np.abs(np.random.normal(0.004, 0.002, n))), 'low': eth_prices_adj * (1 - np.abs(np.random.normal(0.004, 0.002, n))), 'close': eth_prices_adj, 'volume': volumes, }) return df def generate_altcoin_history(btc_df: pd.DataFrame, symbol: str, data_start: str) -> pd.DataFrame: """ 基于BTC走势生成山寨币走势 山寨 = BTC × 随机beta(1.5-2.5) + 额外噪声 """ seed = abs(hash(symbol)) % 2**31 np.random.seed(seed) start_ts = pd.Timestamp(data_start, tz='UTC') mask = btc_df['timestamp'] >= start_ts btc_sub = btc_df[mask].reset_index(drop=True) n = len(btc_sub) if n < 100: return pd.DataFrame() btc_prices = btc_sub['close'].values # 山寨相对BTC的价格比从一个随机起点开始 base_price = np.random.uniform(0.5, 500) ratio = base_price / btc_prices[0] # Beta = 1.5~2.5 beta = np.random.uniform(1.3, 2.0) alt_prices = np.zeros(n) alt_prices[0] = base_price btc_returns = np.diff(np.log(btc_prices)) for i in range(1, n): btc_ret = btc_returns[i-1] if i-1 < len(btc_returns) else 0 alt_ret = btc_ret * beta + np.random.normal(0, 0.004) alt_prices[i] = alt_prices[i-1] * np.exp(alt_ret) volumes = _add_volume_pattern(n, 300, [0.3, 0.6]) np.random.seed(seed + 1) df = pd.DataFrame({ 'timestamp': btc_sub['timestamp'].values, 'open': alt_prices * (1 + np.random.normal(0, 0.001, n)), 'high': alt_prices * (1 + np.abs(np.random.normal(0.005, 0.003, n))), 'low': alt_prices * (1 - np.abs(np.random.normal(0.005, 0.003, n))), 'close': alt_prices, 'volume': volumes, }) return df def generate_all_test_data() -> dict: """ 生成所有12币种的模拟历史数据 Returns: {symbol: df_1h} """ from langlang_ai.config.coins import COINS print("生成模拟历史数据...") btc_df = generate_btc_history() print(f" BTC: {len(btc_df)} bars, " f"{btc_df['timestamp'].min().date()} ~ {btc_df['timestamp'].max().date()}, " f"${btc_df['close'].iloc[0]:.0f} → ${btc_df['close'].iloc[-1]:.0f}") eth_df = generate_eth_history(btc_df) print(f" ETH: {len(eth_df)} bars, " f"${eth_df['close'].iloc[0]:.0f} → ${eth_df['close'].iloc[-1]:.0f}") all_data = {'BTCUSDT': btc_df, 'ETHUSDT': eth_df} # 山寨币 for symbol, cfg in COINS.items(): if symbol in ('BTCUSDT', 'ETHUSDT'): continue alt_df = generate_altcoin_history(btc_df, symbol, cfg.data_start) if len(alt_df) > 0: all_data[symbol] = alt_df print(f" {symbol}: {len(alt_df)} bars, " f"${alt_df['close'].iloc[0]:.2f} → ${alt_df['close'].iloc[-1]:.2f}") return all_data