【実践データ】本記事で使うテストデータのダウンロードについて

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
from scipy.signal import sawtooth, square

def generate_sensor_data(t, waveform='sin', amplitude=1, phase=0, frequency=1, noise_level=0, noise_frequency=0):
    """
    指定されたパラメータに基づいてセンサーデータを生成する関数
    Parameters:
    t (numpy array): 時間データ
    waveform (str): 波形の種類 ('sin', 'cos', 'sawtooth', 'square')
    amplitude (float): 波形の高さ
    phase (float): 波形の位相
    frequency (float): 波形の周波数
    noise_level (float): ノイズの標準偏差
    noise_frequency (float): ノイズの周波数
    Returns:
    numpy array: 生成されたセンサーデータ
    """
    if waveform == 'sin':
        data = amplitude * np.sin(2 * np.pi * frequency * t + phase)
    elif waveform == 'cos':
        data = amplitude * np.cos(2 * np.pi * frequency * t + phase)
    elif waveform == 'sawtooth':
        data = amplitude * sawtooth(2 * np.pi * frequency * t + phase)
    elif waveform == 'square':
        data = amplitude * square(2 * np.pi * frequency * t + phase)
    else:
        raise ValueError("Unsupported waveform type. Use 'sin', 'cos', 'sawtooth', or 'square'.")
    if noise_frequency > 0:
        noise = noise_level * np.sin(2 * np.pi * noise_frequency * t) + np.random.normal(0, noise_level, len(t))
    else:
        noise = np.random.normal(0, noise_level, len(t))
    return data + noise

# 使用例
# サンプル数
num_samples = 1000
t = np.linspace(0, 10, num_samples)
# センサーA: sinカーブ
sensor_A = generate_sensor_data(t, waveform='sin', amplitude=1, phase=0, frequency=1, noise_level=0.4, noise_frequency=0.5)
# センサーB: 0～10を2倍の周期で繰り返すのこぎり刃 + ノイズ
sensor_B = generate_sensor_data(t, waveform='sawtooth', amplitude=2, phase=0, frequency=5, noise_level=0.1, noise_frequency=0.2)
# センサーC: センサーAの3分の1の周期で、少し位相がずれたcos波
sensor_C = generate_sensor_data(t, waveform='cos', amplitude=1, phase=0.5, frequency=15, noise_level=0.1, noise_frequency=0.1)
# センサーD: 5秒ごとに0と1＋ランダム値0～2のsquare波 + ノイズ
sensor_D = generate_sensor_data(t, waveform='sin', amplitude=1, phase=0, frequency=1/2, noise_level=2, noise_frequency=0.1) \
           + generate_sensor_data(t, waveform='sawtooth', amplitude=1, phase=10, frequency=1/5, noise_level=1, noise_frequency=0.4)

sensor_E = generate_sensor_data(t, waveform='cos', amplitude=1, phase=0, frequency=1/3, noise_level=1, noise_frequency=0.4) \
           + generate_sensor_data(t, waveform='square', amplitude=1, phase=10, frequency=1/5, noise_level=1, noise_frequency=0.4)

sensor_F = generate_sensor_data(t, waveform='square', amplitude=2, phase=50, frequency=1, noise_level=0.1, noise_frequency=0.1) \
          + generate_sensor_data(t, waveform='sawtooth', amplitude=1, phase=10, frequency=1/5, noise_level=0.1, noise_frequency=0.1)

sensor_G = generate_sensor_data(t, waveform='sin', amplitude=1, phase=20, frequency=1/3, noise_level=0.1, noise_frequency=5) \
         + generate_sensor_data(t, waveform='sin', amplitude=1, phase=5, frequency=1/4, noise_level=0.1, noise_frequency=12) \
         + generate_sensor_data(t, waveform='sin', amplitude=1, phase=0, frequency=1/5, noise_level=0.1, noise_frequency=3)

sensor_H = generate_sensor_data(t, waveform='sin', amplitude=1, phase=0, frequency=1/5, noise_level=1, noise_frequency=0.2) \
         + generate_sensor_data(t, waveform='sin', amplitude=1, phase=10, frequency=1/2, noise_level=1, noise_frequency=0.3)

sensor_I = generate_sensor_data(t, waveform='sin', amplitude=1, phase=0, frequency=12, noise_level=0.5, noise_frequency=0.2) \

# センサーIの値に異常値を追加（センサーA、E、Gが指定した値を超えた場合、センサーIに異常値（１５）を設定
for i in range(num_samples):
    if sensor_A[i] >= 1.5 and sensor_E[i] < 0 and sensor_G[i] < 0:
        sensor_I[i] = 15

# データフレームにまとめる
data = pd.DataFrame({
    'Sensor_A': sensor_A,
    'Sensor_B': sensor_B,
    'Sensor_C': sensor_C,
    'Sensor_D': sensor_D,
    'Sensor_E': sensor_E,
    'Sensor_F': sensor_F,
    'Sensor_G': sensor_G,
    'Sensor_H': sensor_H,
    'Sensor_I': sensor_I
})


data.to_csv("d:/AnomalyDetectionSample2.csv",encoding="shift-jis",index=None)
fig = GridFig(5, 2, figsize=(15, 9))
fig.add(0, 0, data,data.columns)
fig.show()