# 传统测量误差分析示例
import numpy as np
def traditional_error(temperature, humidity):
    # 模拟环境因素对测量精度的影响
    temp_coeff = 0.05  # 温度系数
    hum_coeff = 0.03   # 湿度系数
    base_error = 0.1   # 基础误差
    return base_error + temp_coeff*abs(temperature-20) + hum_coeff*humidity

print(f"环境温度25°C,湿度60%时误差: {traditional_error(25,60):.2%}")

import tensorflow as tf
from tensorflow.keras import layers

def build_model(input_shape):
    inputs = layers.Input(shape=input_shape)

    # CNN分支处理空间特征
    cnn_branch = layers.Conv1D(64, 3, activation='relu')(inputs)
    cnn_branch = layers.MaxPooling1D(2)(cnn_branch)

    # Transformer分支处理时序依赖
    transformer_branch = layers.MultiHeadAttention(num_heads=4, key_dim=32)(inputs, inputs)
    transformer_branch = layers.LayerNormalization(epsilon=1e-6)(transformer_branch)

    # 特征融合与输出
    merged = layers.Concatenate()([cnn_branch, transformer_branch])
    outputs = layers.Dense(1, activation='linear')(merged)

    model = tf.keras.Model(inputs=inputs, outputs=outputs)
    model.compile(optimizer='adam', loss='mse')
    return model

model = build_model((100, 8))  # 输入形状：序列长度100，特征维度8
model.summary()

def dynamic_threshold(history_data, window_size=24):
    model = Sequential()
    model.add(LSTM(50, input_shape=(window_size, history_data.shape[1])))
    model.add(Dense(1))
    model.compile(optimizer='adam', loss='mse')

    # 训练模型预测正常状态阈值
    X, y = create_dataset(history_data, window_size)
    model.fit(X, y, epochs=50, batch_size=32)

    # 计算当前数据与预测值的偏差
    current_pred = model.predict(current_data)
    deviation = np.abs(current_data - current_pred)
    return deviation > threshold

# 使用预训练ResNet50进行迁移学习
base_model = tf.keras.applications.ResNet50(weights='imagenet', include_top=False)
x = base_model.output
x = GlobalAveragePooling2D()(x)
x = Dense(1024, activation='relu')(x)
predictions = Dense(num_classes, activation='softmax')(x)

model = Model(inputs=base_model.input, outputs=predictions)

# 冻结底层卷积层
for layer in base_model.layers:
    layer.trainable = False

model.compile(optimizer='adam', loss='categorical_crossentropy')