377 lines
13 KiB
Python
377 lines
13 KiB
Python
"""
|
||
EMA趋势策略参数优化扫描
|
||
|
||
基于前一轮回测发现 EMA-Trend 方向盈利 +327 USDT，仅差 101 USDT 即可盈利。
|
||
核心优化方向：减少交易次数（降低费用），同时保持方向盈利。
|
||
|
||
扫描参数：
|
||
- fast_ema: [8, 13, 15, 20]
|
||
- slow_ema: [21, 34, 40, 55]
|
||
- big_ema: [120, 200, 300]
|
||
- atr_min_pct: [0.0003, 0.0005, 0.0008, 0.0012]
|
||
- stop_loss: [0.003, 0.004, 0.005, 0.006]
|
||
- max_hold: [900, 1200, 1800, 2700, 3600]
|
||
"""
|
||
|
||
import time
|
||
import datetime
|
||
import sqlite3
|
||
import itertools
|
||
from pathlib import Path
|
||
from dataclasses import dataclass
|
||
from typing import List
|
||
|
||
|
||
# ========================= EMA =========================
|
||
class EMA:
|
||
def __init__(self, period):
|
||
self.k = 2.0 / (period + 1)
|
||
self.value = None
|
||
|
||
def update(self, price):
|
||
if self.value is None:
|
||
self.value = price
|
||
else:
|
||
self.value = price * self.k + self.value * (1 - self.k)
|
||
return self.value
|
||
|
||
|
||
# ========================= 数据加载 =========================
|
||
def load_data(start_date='2025-01-01', end_date='2025-12-31'):
|
||
db_path = Path(__file__).parent.parent / 'models' / 'database.db'
|
||
start_ms = int(datetime.datetime.strptime(start_date, '%Y-%m-%d').timestamp()) * 1000
|
||
end_ms = int((datetime.datetime.strptime(end_date, '%Y-%m-%d') + datetime.timedelta(days=1)).timestamp()) * 1000
|
||
|
||
conn = sqlite3.connect(str(db_path))
|
||
rows = conn.cursor().execute(
|
||
"SELECT id, open, high, low, close FROM bitmart_eth_1m WHERE id >= ? AND id < ? ORDER BY id",
|
||
(start_ms, end_ms)
|
||
).fetchall()
|
||
conn.close()
|
||
|
||
data = []
|
||
for r in rows:
|
||
data.append({
|
||
'datetime': datetime.datetime.fromtimestamp(r[0] / 1000.0),
|
||
'open': r[1], 'high': r[2], 'low': r[3], 'close': r[4],
|
||
})
|
||
return data
|
||
|
||
|
||
# ========================= 快速回测引擎 =========================
|
||
def run_ema_backtest(data, fast_p, slow_p, big_p, atr_period, atr_min,
|
||
stop_loss_pct, hard_sl_pct, max_hold_sec,
|
||
initial_balance=1000.0, leverage=50, risk_pct=0.005,
|
||
taker_fee=0.0006, rebate_rate=0.90, min_hold_sec=200):
|
||
"""
|
||
快速回测 EMA 趋势策略，返回关键指标字典
|
||
"""
|
||
balance = initial_balance
|
||
position = 0 # -1/0/1
|
||
open_price = 0.0
|
||
open_time = None
|
||
pos_size = 0.0
|
||
pending = None
|
||
|
||
ema_f = EMA(fast_p)
|
||
ema_s = EMA(slow_p)
|
||
ema_b = EMA(big_p)
|
||
|
||
highs = []
|
||
lows = []
|
||
closes = []
|
||
prev_fast = None
|
||
prev_slow = None
|
||
|
||
trade_count = 0
|
||
win_count = 0
|
||
total_dir_pnl = 0.0
|
||
total_fee = 0.0
|
||
total_rebate = 0.0
|
||
|
||
def calc_atr():
|
||
if len(highs) < atr_period + 1:
|
||
return None
|
||
trs = []
|
||
for i in range(-atr_period, 0):
|
||
tr = max(highs[i] - lows[i],
|
||
abs(highs[i] - closes[i-1]),
|
||
abs(lows[i] - closes[i-1]))
|
||
trs.append(tr)
|
||
return sum(trs) / len(trs)
|
||
|
||
def do_open(direction, price, dt_):
|
||
nonlocal balance, position, open_price, open_time, pos_size, total_fee
|
||
ps = balance * risk_pct * leverage
|
||
if ps < 1:
|
||
return
|
||
fee_ = ps * taker_fee
|
||
balance -= fee_
|
||
position = 1 if direction == 'long' else -1
|
||
open_price = price
|
||
open_time = dt_
|
||
pos_size = ps
|
||
|
||
def do_close(price, dt_):
|
||
nonlocal balance, position, open_price, open_time, pos_size
|
||
nonlocal trade_count, win_count, total_dir_pnl, total_fee, total_rebate, pending
|
||
if position == 0:
|
||
return
|
||
if position == 1:
|
||
pp = (price - open_price) / open_price
|
||
else:
|
||
pp = (open_price - price) / open_price
|
||
|
||
pnl_ = pos_size * pp
|
||
cv = pos_size * (1 + pp)
|
||
cf = cv * taker_fee
|
||
of = pos_size * taker_fee
|
||
tf = of + cf
|
||
rb = tf * rebate_rate
|
||
|
||
balance += pnl_ - cf + rb
|
||
total_dir_pnl += pnl_
|
||
total_fee += tf
|
||
total_rebate += rb
|
||
trade_count += 1
|
||
if pnl_ > 0:
|
||
win_count += 1
|
||
|
||
position = 0
|
||
open_price = 0.0
|
||
open_time = None
|
||
pos_size = 0.0
|
||
pending = None
|
||
|
||
for bar in data:
|
||
price = bar['close']
|
||
dt = bar['datetime']
|
||
|
||
highs.append(bar['high'])
|
||
lows.append(bar['low'])
|
||
closes.append(price)
|
||
|
||
fast = ema_f.update(price)
|
||
slow = ema_s.update(price)
|
||
big = ema_b.update(price)
|
||
|
||
atr = calc_atr()
|
||
atr_pct = atr / price if atr and price > 0 else 0
|
||
|
||
cross_up = (prev_fast is not None and prev_fast <= prev_slow and fast > slow)
|
||
cross_down = (prev_fast is not None and prev_fast >= prev_slow and fast < slow)
|
||
prev_fast = fast
|
||
prev_slow = slow
|
||
|
||
# === 有持仓 ===
|
||
if position != 0 and open_time:
|
||
if position == 1:
|
||
p = (price - open_price) / open_price
|
||
else:
|
||
p = (open_price - price) / open_price
|
||
hs = (dt - open_time).total_seconds()
|
||
|
||
# 硬止损
|
||
if -p >= hard_sl_pct:
|
||
do_close(price, dt)
|
||
continue
|
||
|
||
can_close_ = hs >= min_hold_sec
|
||
|
||
if can_close_:
|
||
# 止损
|
||
if -p >= stop_loss_pct:
|
||
do_close(price, dt)
|
||
continue
|
||
# 超时
|
||
if hs >= max_hold_sec:
|
||
do_close(price, dt)
|
||
continue
|
||
# 反手
|
||
if position == 1 and cross_down:
|
||
do_close(price, dt)
|
||
if price < big and atr_pct >= atr_min:
|
||
do_open('short', price, dt)
|
||
continue
|
||
if position == -1 and cross_up:
|
||
do_close(price, dt)
|
||
if price > big and atr_pct >= atr_min:
|
||
do_open('long', price, dt)
|
||
continue
|
||
# 延迟信号
|
||
if pending == 'close_long' and position == 1:
|
||
do_close(price, dt)
|
||
if fast < slow and price < big and atr_pct >= atr_min:
|
||
do_open('short', price, dt)
|
||
continue
|
||
if pending == 'close_short' and position == -1:
|
||
do_close(price, dt)
|
||
if fast > slow and price > big and atr_pct >= atr_min:
|
||
do_open('long', price, dt)
|
||
continue
|
||
else:
|
||
if position == 1 and cross_down:
|
||
pending = 'close_long'
|
||
elif position == -1 and cross_up:
|
||
pending = 'close_short'
|
||
|
||
# === 无持仓 ===
|
||
if position == 0 and atr_pct >= atr_min:
|
||
if cross_up and price > big:
|
||
do_open('long', price, dt)
|
||
elif cross_down and price < big:
|
||
do_open('short', price, dt)
|
||
|
||
# 强制平仓
|
||
if position != 0:
|
||
do_close(data[-1]['close'], data[-1]['datetime'])
|
||
|
||
net = balance - initial_balance
|
||
net_fee_cost = total_fee - total_rebate
|
||
vol = total_fee / taker_fee if taker_fee > 0 else 0
|
||
wr = win_count / trade_count * 100 if trade_count > 0 else 0
|
||
avg_dir = total_dir_pnl / trade_count if trade_count > 0 else 0
|
||
|
||
return {
|
||
'balance': balance,
|
||
'net': net,
|
||
'net_pct': net / initial_balance * 100,
|
||
'trades': trade_count,
|
||
'win_rate': wr,
|
||
'dir_pnl': total_dir_pnl,
|
||
'total_fee': total_fee,
|
||
'rebate': total_rebate,
|
||
'net_fee': net_fee_cost,
|
||
'volume': vol,
|
||
'avg_dir_pnl': avg_dir,
|
||
}
|
||
|
||
|
||
# ========================= 参数扫描 =========================
|
||
def main():
|
||
print("Loading data...")
|
||
data = load_data('2025-01-01', '2025-12-31')
|
||
print(f"Loaded {len(data)} bars\n")
|
||
|
||
# 参数组合
|
||
param_grid = {
|
||
'fast_p': [8, 13, 15, 20],
|
||
'slow_p': [21, 34, 40, 55],
|
||
'big_p': [120, 200, 300],
|
||
'atr_min': [0.0003, 0.0005, 0.0008, 0.0012],
|
||
'stop_loss_pct':[0.003, 0.004, 0.005, 0.008],
|
||
'max_hold_sec': [900, 1200, 1800, 3600],
|
||
}
|
||
|
||
# 过滤无效组合 (fast >= slow)
|
||
combos = []
|
||
for fp, sp, bp, am, sl, mh in itertools.product(
|
||
param_grid['fast_p'], param_grid['slow_p'], param_grid['big_p'],
|
||
param_grid['atr_min'], param_grid['stop_loss_pct'], param_grid['max_hold_sec']
|
||
):
|
||
if fp >= sp:
|
||
continue
|
||
combos.append((fp, sp, bp, am, sl, mh))
|
||
|
||
print(f"Total parameter combinations: {len(combos)}")
|
||
print(f"Estimated time: ~{len(combos) * 2 / 60:.0f} minutes\n")
|
||
|
||
# 只跑最有潜力的子集（减少扫描时间）
|
||
# 基于前次回测，聚焦在更长周期EMA（减少交易）+ 更高ATR过滤（质量过滤）
|
||
focused_combos = []
|
||
for fp, sp, bp, am, sl, mh in combos:
|
||
# 过滤：聚焦减少交易次数的参数
|
||
if fp < 8:
|
||
continue
|
||
if sp < 21:
|
||
continue
|
||
focused_combos.append((fp, sp, bp, am, sl, mh))
|
||
|
||
print(f"Focused combinations: {len(focused_combos)}")
|
||
|
||
# 如果组合太多，进一步采样
|
||
if len(focused_combos) > 500:
|
||
# 分两轮：先粗扫，再精调
|
||
print("Phase 1: Coarse scan with subset...")
|
||
coarse_combos = []
|
||
for fp, sp, bp, am, sl, mh in focused_combos:
|
||
if am in [0.0003, 0.0008] and sl in [0.004, 0.006] and mh in [1200, 1800]:
|
||
coarse_combos.append((fp, sp, bp, am, sl, mh))
|
||
elif am in [0.0005, 0.0012] and sl in [0.003, 0.005, 0.008] and mh in [900, 3600]:
|
||
coarse_combos.append((fp, sp, bp, am, sl, mh))
|
||
focused_combos = coarse_combos[:600] # cap
|
||
print(f" Reduced to {len(focused_combos)} combos")
|
||
|
||
results = []
|
||
t0 = time.time()
|
||
|
||
for idx, (fp, sp, bp, am, sl, mh) in enumerate(focused_combos):
|
||
r = run_ema_backtest(
|
||
data, fast_p=fp, slow_p=sp, big_p=bp,
|
||
atr_period=14, atr_min=am,
|
||
stop_loss_pct=sl, hard_sl_pct=sl * 1.5,
|
||
max_hold_sec=mh,
|
||
)
|
||
r['params'] = f"EMA({fp}/{sp}/{bp}) ATR>{am*100:.2f}% SL={sl*100:.1f}% MaxH={mh}s"
|
||
r['fp'] = fp
|
||
r['sp'] = sp
|
||
r['bp'] = bp
|
||
r['am'] = am
|
||
r['sl'] = sl
|
||
r['mh'] = mh
|
||
results.append(r)
|
||
|
||
if (idx + 1) % 50 == 0:
|
||
elapsed = time.time() - t0
|
||
eta = elapsed / (idx + 1) * (len(focused_combos) - idx - 1)
|
||
print(f" [{idx+1}/{len(focused_combos)}] elapsed={elapsed:.0f}s eta={eta:.0f}s")
|
||
|
||
total_time = time.time() - t0
|
||
print(f"\nScan complete! {len(results)} combos in {total_time:.1f}s")
|
||
|
||
# 按净收益排序
|
||
results.sort(key=lambda x: x['net'], reverse=True)
|
||
|
||
# === 打印 Top 20 ===
|
||
print(f"\n{'='*120}")
|
||
print(f" TOP 20 PARAMETER COMBINATIONS (by Net P&L)")
|
||
print(f"{'='*120}")
|
||
print(f" {'#':>3} {'Params':<52} {'Net%':>7} {'Net$':>9} {'Trades':>7} {'WR':>6} {'DirPnL':>9} {'Rebate':>9} {'NetFee':>8}")
|
||
print(f" {'-'*116}")
|
||
|
||
for i, r in enumerate(results[:20]):
|
||
print(f" {i+1:>3} {r['params']:<52} {r['net_pct']:>+6.2f}% {r['net']:>+8.2f} {r['trades']:>7} {r['win_rate']:>5.1f}% {r['dir_pnl']:>+8.2f} {r['rebate']:>8.2f} {r['net_fee']:>8.2f}")
|
||
|
||
# === 打印 Bottom 5 (最差) ===
|
||
print(f"\n BOTTOM 5:")
|
||
print(f" {'-'*116}")
|
||
for i, r in enumerate(results[-5:]):
|
||
print(f" {len(results)-4+i:>3} {r['params']:<52} {r['net_pct']:>+6.2f}% {r['net']:>+8.2f} {r['trades']:>7} {r['win_rate']:>5.1f}% {r['dir_pnl']:>+8.2f} {r['rebate']:>8.2f} {r['net_fee']:>8.2f}")
|
||
|
||
print(f"{'='*120}")
|
||
|
||
# === 盈利的参数组合统计 ===
|
||
profitable = [r for r in results if r['net'] > 0]
|
||
print(f"\nProfitable combinations: {len(profitable)} / {len(results)} ({len(profitable)/len(results)*100:.1f}%)")
|
||
|
||
if profitable:
|
||
print(f"\nAll profitable combinations:")
|
||
print(f" {'#':>3} {'Params':<52} {'Net%':>7} {'Net$':>9} {'Trades':>7} {'WR':>6} {'DirPnL':>9} {'Rebate':>9}")
|
||
print(f" {'-'*106}")
|
||
for i, r in enumerate(profitable):
|
||
print(f" {i+1:>3} {r['params']:<52} {r['net_pct']:>+6.2f}% {r['net']:>+8.2f} {r['trades']:>7} {r['win_rate']:>5.1f}% {r['dir_pnl']:>+8.2f} {r['rebate']:>8.2f}")
|
||
|
||
# 保存全部结果到CSV
|
||
csv_path = Path(__file__).parent.parent / 'param_scan_results.csv'
|
||
with open(csv_path, 'w', encoding='utf-8-sig') as f:
|
||
f.write("fast,slow,big,atr_min,stop_loss,max_hold,net_pct,net_usd,trades,win_rate,dir_pnl,rebate,net_fee,volume\n")
|
||
for r in results:
|
||
f.write(f"{r['fp']},{r['sp']},{r['bp']},{r['am']},{r['sl']},{r['mh']},"
|
||
f"{r['net_pct']:.4f},{r['net']:.4f},{r['trades']},{r['win_rate']:.2f},"
|
||
f"{r['dir_pnl']:.4f},{r['rebate']:.4f},{r['net_fee']:.4f},{r['volume']:.0f}\n")
|
||
print(f"\nFull results saved: {csv_path}")
|
||
|
||
|
||
if __name__ == '__main__':
|
||
main()
|