1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
|
class MultiFactorDemandForecaster:
def __init__(self):
self.models = {}
self.feature_importance = {}
def prepare_multi_factor_features(self, data):
"""多因素特征工程"""
df = data.copy()
df['year'] = df['date'].dt.year
df['month'] = df['date'].dt.month
df['day'] = df['date'].dt.day
df['dayofweek'] = df['date'].dt.dayofweek
df['is_weekend'] = df['dayofweek'].isin([5, 6]).astype(int)
df['temperature_ma7'] = df['temperature'].rolling(7).mean()
df['rainfall_ma7'] = df['rainfall'].rolling(7).mean()
df['weather_score'] = df['temperature'] / 30 - df['rainfall'] / 10
df['promotion_intensity'] = df['promotion_discount'] * df['promotion_duration']
df['is_promotion'] = (df['promotion_discount'] > 0).astype(int)
df['competitor_price_ratio'] = df['competitor_price'] / df['our_price']
df['price_advantage'] = (df['competitor_price'] - df['our_price']) / df['our_price']
df['gdp_growth_ma30'] = df['gdp_growth'].rolling(30).mean()
df['inflation_ma30'] = df['inflation'].rolling(30).mean()
df['social_mention_ma7'] = df['social_mention'].rolling(7).mean()
df['sentiment_score_ma7'] = df['sentiment_score'].rolling(7).mean()
for lag in [1, 7, 14, 30]:
df[f'demand_lag_{lag}'] = df['demand'].shift(lag)
df[f'price_lag_{lag}'] = df['our_price'].shift(lag)
df['price_weather_interaction'] = df['our_price'] * df['weather_score']
df['promotion_weather_interaction'] = df['promotion_intensity'] * df['weather_score']
self.feature_columns = [col for col in df.columns
if col not in ['date', 'demand']]
return df
def train_ensemble_model(self, data):
"""训练集成模型"""
df = self.prepare_multi_factor_features(data)
df = df.dropna()
X = df[self.feature_columns]
y = df['demand']
models = {
'random_forest': RandomForestRegressor(n_estimators=100, random_state=42),
'linear': LinearRegression(),
'gradient_boosting': GradientBoostingRegressor(n_estimators=100, random_state=42)
}
for name, model in models.items():
model.fit(X, y)
self.models[name] = model
if hasattr(model, 'feature_importances_'):
self.feature_importance[name] = dict(zip(self.feature_columns,
model.feature_importances_))
self.ensemble_weights = self.calculate_ensemble_weights(data)
def calculate_ensemble_weights(self, data):
"""计算集成权重"""
df = self.prepare_multi_factor_features(data)
df = df.dropna()
split_idx = int(len(df) * 0.8)
train_data = df[:split_idx]
test_data = df[split_idx:]
X_test = test_data[self.feature_columns]
y_test = test_data['demand']
predictions = {}
for name, model in self.models.items():
pred = model.predict(X_test)
predictions[name] = pred
from scipy.optimize import minimize
def objective(weights):
ensemble_pred = sum(weights[i] * predictions[name]
for i, name in enumerate(self.models.keys()))
return np.mean((y_test - ensemble_pred) ** 2)
constraints = {'type': 'eq', 'fun': lambda w: np.sum(w) - 1}
bounds = [(0, 1) for _ in self.models.keys()]
result = minimize(objective, [1/len(self.models)] * len(self.models),
method='SLSQP', bounds=bounds, constraints=constraints)
return dict(zip(self.models.keys(), result.x))
def predict_ensemble(self, data, horizon=7):
"""集成预测"""
predictions = []
current_data = data.copy()
for _ in range(horizon):
df = self.prepare_multi_factor_features(current_data)
latest_features = df[self.feature_columns].iloc[-1:].values
model_predictions = {}
for name, model in self.models.items():
pred = model.predict(latest_features)[0]
model_predictions[name] = pred
ensemble_pred = sum(self.ensemble_weights[name] * pred
for name, pred in model_predictions.items())
predictions.append(ensemble_pred)
new_date = current_data['date'].iloc[-1] + pd.Timedelta(days=1)
new_row = pd.DataFrame({
'date': [new_date],
'demand': [ensemble_pred]
})
current_data = pd.concat([current_data, new_row], ignore_index=True)
return predictions
def analyze_feature_importance(self):
"""分析特征重要性"""
importance_df = pd.DataFrame(self.feature_importance)
importance_df['average'] = importance_df.mean(axis=1)
importance_df = importance_df.sort_values('average', ascending=False)
return importance_df
np.random.seed(42)
n_days = 365
dates = pd.date_range('2023-01-01', periods=n_days, freq='D')
data = pd.DataFrame({
'date': dates,
'demand': 100 + 20 * np.sin(2 * np.pi * np.arange(n_days) / 365) + np.random.normal(0, 10, n_days),
'temperature': 20 + 10 * np.sin(2 * np.pi * np.arange(n_days) / 365) + np.random.normal(0, 3, n_days),
'rainfall': np.random.exponential(2, n_days),
'promotion_discount': np.random.choice([0, 0.1, 0.2, 0.3], n_days, p=[0.7, 0.15, 0.1, 0.05]),
'promotion_duration': np.random.choice([0, 3, 7, 14], n_days, p=[0.7, 0.15, 0.1, 0.05]),
'our_price': 100 + np.random.normal(0, 5, n_days),
'competitor_price': 105 + np.random.normal(0, 5, n_days),
'gdp_growth': np.random.normal(3, 1, n_days),
'inflation': np.random.normal(2, 0.5, n_days),
'social_mention': np.random.poisson(10, n_days),
'sentiment_score': np.random.normal(0.5, 0.2, n_days)
})
multi_forecaster = MultiFactorDemandForecaster()
multi_forecaster.train_ensemble_model(data)
ensemble_predictions = multi_forecaster.predict_ensemble(data, 7)
print("集成预测结果:", ensemble_predictions)
importance = multi_forecaster.analyze_feature_importance()
print("特征重要性分析:")
print(importance.head(10))
|
微信
