### BP神经网络预测航班滑出时间 为了实现基于BP神经网络的航班滑出时间预测，在Jupyter Notebook环境中编写代码可以分为几个部分完成。下面提供了一个完整的Python代码示例，用于构建并训练一个简单的前馈神经网络来进行此任务。 ```python import numpy as np from mxnet import nd, autograd, gluon from mxnet.gluon import nn from sklearn.model_selection import train_test_split from sklearn.preprocessing import StandardScaler import pandas as pd # 加载数据集（这里假设有一个CSV文件作为输入） data = pd.read_csv('flight_data.csv') # 数据预处理 features = data.drop(['departure_delay'], axis=1).values # 输入特征 labels = data['departure_delay'].values.reshape(-1, 1) # 输出标签 scaler_features = StandardScaler() scaler_labels = StandardScaler() scaled_features = scaler_features.fit_transform(features) scaled_labels = scaler_labels.fit_transform(labels) X_train, X_test, y_train, y_test = train_test_split(scaled_features, scaled_labels, test_size=0.2, random_state=42) # 将NumPy数组转换成NDArray形式 train_data = gluon.data.ArrayDataset(nd.array(X_train), nd.array(y_train)) test_data = gluon.data.ArrayDataset(nd.array(X_test)) batch_size = 64 train_loader = gluon.data.DataLoader(train_data, batch_size=batch_size, shuffle=True) test_loader = gluon.data.DataLoader(test_data, batch_size=len(test_data)) # 构建模型 net = nn.Sequential() with net.name_scope(): net.add(nn.Dense(32, activation='relu')) # 添加一层带有ReLU激活函数的全连接层 net.add(nn.Dense(16, activation='relu')) net.add(nn.Dense(1)) # 输出层只有一个节点 ctx = mx.cpu() # 使用CPU上下文环境 net.initialize(ctx=ctx) # 定义损失函数和优化器 loss_function = gluon.loss.L2Loss() # 平方误差损失 trainer = gluon.Trainer(net.collect_params(), 'adam', {'learning_rate': 0.01}) epochs = 50 # 训练轮数设置为50次 for e in range(epochs): cumulative_loss = 0 for i, (data, label) in enumerate(train_loader): data = data.as_in_context(ctx) label = label.as_in_context(ctx) with autograd.record(): # 开始记录梯度计算图 output = net(data) loss = loss_function(output, label) loss.backward() # 反向传播求导 trainer.step(batch_size) # 更新参数 cumulative_loss += nd.mean(loss).asscalar() print(f"Epoch {e}, Loss: {cumulative_loss / len(train_loader)}") # 测试阶段 predictions = [] for data in test_loader: data = data.as_in_context(ctx) prediction = net(data).asnumpy().flatten() predictions.extend(prediction.tolist()) predicted_delays = scaler_labels.inverse_transform(np.array(predictions).reshape(-1, 1)).flatten() actual_delays = scaler_labels.inverse_transform(y_test).flatten() mse = ((predicted_delays - actual_delays)**2).mean(axis=None) print(f'Mean Squared Error on Test Set: {mse}') ``` 上述代码展示了如何利用MXNet框架创建一个多层感知机(Multilayer Perceptron)，并通过反向传播算法调整权重以最小化均方误差[MSE]。该过程涉及到了数据加载、标准化处理以及评估最终性能指标等方面的工作[^2][^4]。