SRN对时间序列进行预测

1.导包（安装好tensorflow2）

import tensorflow as tf
from matplotlib import pyplot as plt
import numpy as np
from keras.models import Sequential
from keras.layers import Dense, SimpleRNN
from sklearn.model_selection import train_test_split
import pandas as pd

2.设置时间步和用于存储损失和未来预测的数据列表

time_step = 7
loss_metrics_data = []
future_predictions_data = []

3.从序列数据中提取输入特征和目标

def extract_data(data, time_step):
    X = []
    y = []
    for i in range(len(data) - time_step):
        X.append([a for a in data[i:i + time_step]])
        y.append(data[i + time_step])
    X = np.array(X)
    X = X.reshape(X.shape[0], X.shape[1], 1)
    return X, y

4.读取数据

df = pd.read_csv("单品编号_售价_销量_品种（完整时间）（异常值处理）.csv")

数据集如下：

5.根据分类名称和销售日期对销量求和

grouped = df.groupby(['分类名称', '销售日期'])['销量(千克)'].sum().reset_index()

6.创建一个字典，每个类别对应一个 DataFrame

dfs = {}
for category in grouped['分类名称'].unique():
    dfs[category] = grouped[grouped['分类名称'] == category]
    y = dfs[category]['销量(千克)']

7.将数据分为训练集和测试集

train, test = train_test_split(y, test_size=0.25, shuffle=False)
    train = train.reset_index(drop=True)
    test = test.reset_index(drop=True)

    max_train = max(train)
    train_norm = train / max_train
    test_norm = test / max_train

8.提取时间序列特征

X_train, y_train = extract_data(train_norm, time_step)
    X_test, y_test = extract_data(test_norm, time_step)

9.绘制训练数据

fig1 = plt.figure(figsize=(8, 5))
    plt.plot(train)
    plt.xlabel("时间")
    plt.ylabel("销量")
    plt.show()

10.创建并训练 Simple RNN 模型

model = Sequential()
    model.add(SimpleRNN(units=8, input_shape=(time_step, 1), activation='relu'))
    model.add(Dense(units=1, activation='linear'))
    model.compile(optimizer='adam', loss="mean_squared_error")
    model.summary()
    model.fit(X_train, y_train, batch_size=128, epochs=300)

11.使用模型进行预测

y_pred = model.predict(X_test)

12.反归一化预测值和测试值

y_pred_original = y_pred * max_train
    y_test_original = np.array(y_test) * max_train

13.绘制真实值和预测值的图形

plt.figure(figsize=(15, 6))
    plt.plot(y_test_original, label="真实值", color='blue')
    plt.plot(y_pred_original, label="预测值", color='red')
    plt.title(f"{category} 的预测")
    plt.xlabel("时间步")
    plt.ylabel("数值")
    plt.legend()
    plt.show()

14.计算损失函数指标

mse_values = []
    loss_test = tf.losses.mean_squared_error(y_test, y_pred).numpy()
    mse_values.append(loss_test)
    average_mse = np.mean(mse_values)
    print(f"{category} 测试集上的损失 (MSE): {average_mse}")

    rmse = np.sqrt(average_mse)
    print(f"{category} 的 RMSE: {rmse}")

    mape = 100 * np.mean(np.abs((y_test_original - y_pred_original) / y_test))
    print(f"{category} 的 MAPE: {mape}%")

    rmspe = 100 * np.sqrt(np.mean(((y_test_original - y_pred_original) / y_test) ** 2))
    print(f"{category} 的 RMSPE: {rmspe}%")

    mae = np.mean(np.abs(y_test_original - y_pred_original))
    print(f"{category} 的 MAE: {mae}")

    naive_forecast_error = np.mean(np.abs(y_test_original[1:] - y_test_original[:-1]))
    mase = mae / naive_forecast_error
    print(f"{category} 的 MASE: {mase}")

15.获取未来7天的预测

last_data = np.array(test_norm[-time_step:]).reshape(1, time_step, 1)
    future_predictions = []
    for _ in range(7):
        future_pred = model.predict(last_data)
        future_predictions.append(future_pred[0][0])
        last_data = np.append(last_data[0][1:], future_pred).reshape(1, time_step, 1)

16.反归一化未来预测值

future_predictions = np.array(future_predictions) * max_train
    print(f"{category} 的未来7天预测: {future_predictions}")