基于高斯混合模型聚类CNN-BiLSTM-Attention风电场短期功率预测方法（Python和Matlab代码实现）

from __future__ import print_function import argparse import os import matplotlib.pyplot as plt import numpy as np import pandas as pd from model import train_model parser = argparse.ArgumentParser() parser.add_argument('--model_name', type=str, default='CNN_BiLSTM', choices=['CNN_BiLSTM_Attn', 'CNN_BiLSTM']) parser.add_argument('--epochs', type=int, default=600, help="Epochs") parser.add_argument('--batch_size', type=int, default=256, help="Batch Size") parser.add_argument('--learning_rate', type=float, default=0.001, help="Learning rate") parser.add_argument('--sequence_length', type=int, default=10, help="sequence length") parser.add_argument('--dataset', type=str, default='all', choices=['cluster1', 'cluster2', 'cluster3', 'all']) parser.add_argument('--path', type=str, default='./results/') args = parser.parse_args() if not os.path.exists(args.path): os.mkdir(args.path) def convertSeriesToMatrix(vectorSeries, sequence_length): """ 滑动时间窗口处理 """ matrix = [] for i in range(len(vectorSeries) - sequence_length + 1): matrix.append(vectorSeries[i: i + sequence_length]) return matrix if args.dataset in {'cluster1', 'cluster2', 'cluster3'}: df = pd.read_csv(args.dataset + '.csv') wind = np.array(df) elif args.dataset == 'all': X_files = os.listdir(os.getcwd() + '/dataset') if '.gitignore' in X_files: X_files.remove('.gitignore') X_files = np.array(X_files) wind = [] for filename in X_files: df = pd.read_csv(os.getcwd() + '/dataset/' + filename) data = df.values wind.append(data[:, 0]) wind = np.sum(wind, axis=0).reshape(-1, 1) list_hourly_data = [wind[i] for i in range(0, wind.shape[0]) if i % 12 == 0] wind_data = np.array(list_hourly_data) # min-max归一化处理 MAX_MIN = [] for i in range(wind_data.shape[1]): # 取每一列 temp_max = np.max(wind_data[:, i]) temp_min = np.min(wind_data[:, i]) # 第i列最小值 wind_data[:, i] = (wind_data[:, i] - temp_min) / (temp_max - temp_min) # 归一化 # wind_data[:, i] = wind_data[:, i] / temp_max # 归一化 MAX_MIN.append([temp_max, temp_min]) MAX_NUM = MAX_MIN[0][0] MIN_NUM = MAX_MIN[0][1] matrix_load = convertSeriesToMatrix(wind_data, args.sequence_length) # 滑动时间窗口，窗口长度可以改 matrix_load = np.array(matrix_load).astype(np.float32) # matrix_load转为ndarray # 分为训练集与测试集，训练集占90% train_row = int(round(0.9 * matrix_load.shape[0])) train_set = matrix_load[:train_row, :] np.random.shuffle(train_set) # 打乱训练集，测试集不用打乱 X_train = train_set[:, :-1] # 训练集 y_train = train_set[:, -1, 0].reshape(-1, 1) # 训练集label X_test = matrix_load[train_row:, :-1] # 测试集 y_test = matrix_load[train_row:, -1, 0].reshape(-1, 1) # 测试集label X_train = np.reshape(X_train, (X_train.shape[0], X_train.shape[1] * X_train.shape[2], 1)) X_test = np.reshape(X_test, (X_test.shape[0], X_test.shape[1] * X_test.shape[2], 1)) # ===================================== # ========= 载入模型，开始训练 ========= # ===================================== input_shape = X_train.shape[1:] model = train_model(args.model_name, input_shape=input_shape) model.summary() # print model 结构 history = model.fit(X_train, y_train, batch_size=args.batch_size, epochs=args.epochs, validation_split=0.05, verbose=1) # 开始训练 # model.save(args.path + args.model_name + '_' + args.dataset + 'model.h5') # 保存模型 # ===================================== # ============ 训练结果评估 ============ # ===================================== # 预测测试集 predicted_values = model.predict(X_test) num_test_samples = len(predicted_values) predicted_values = np.reshape(predicted_values, (num_test_samples, 1)) # 评估训练结果 test_mse = model.evaluate(X_test, y_test, verbose=1) print('\nThe mean squared error (MSE) on the test data set is %.3f over %d test samples.' % (test_mse[0], len(y_test))) # 反归一化 predicted_values = predicted_values * (MAX_NUM - MIN_NUM) + MIN_NUM y_test = y_test * (MAX_NUM - MIN_NUM) + MIN_NUM y_train = y_train * (MAX_NUM - MIN_NUM) + MIN_NUM # 画图 fig = plt.figure(figsize=(15, 5), dpi=600) plt.plot(y_test) plt.plot(predicted_values) plt.xlabel('Hour') plt.ylabel('Wind Power') # plt.show() fig.savefig(args.path + args.model_name + '_' + args.dataset + '_output.png', bbox_inches='tight') # MSE 损失函数 fig = plt.figure(figsize=(15, 5), dpi=600) plt.plot(history.history['loss'], label='training loss') plt.plot(history.history['val_loss'], label='val loss') plt.title('model loss') plt.ylabel('MSE Loss') plt.xlabel('epoch') plt.legend(loc='upper right') fig.savefig(args.path + args.model_name + '_' + args.dataset + '_loss.png') # MAE fig = plt.figure(figsize=(15, 5), dpi=600) plt.plot(history.history['mae'], label='training mae') plt.plot(history.history['val_mae'], label='val mae') plt.title('model MAE') plt.ylabel('MAE') plt.xlabel('epoch') plt.legend(loc='upper right') fig.savefig(args.path + args.model_name + '_' + args.dataset + '_mae.png') # RMSE fig = plt.figure(figsize=(15, 5), dpi=600) plt.plot(history.history['root_mean_squared_error'], label='training rmse') plt.plot(history.history['val_root_mean_squared_error'], label='val rmse') plt.title('model RMSE') plt.ylabel('RMSE') plt.xlabel('epoch') plt.legend(loc='upper right') fig.savefig(args.path + args.model_name + '_' + args.dataset + '_rmse.png') # 将预测结果保存成csv文件 data = np.hstack((predicted_values, y_test)) df = pd.DataFrame(data, columns=['predicted', 'real']) df.to_csv(args.path + args.model_name + '_' + args.dataset + '_result.csv') # MSE loss 和 MAE 保存成csv文件 df = pd.DataFrame(history.history) df.to_csv(args.path + args.model_name + '_' + args.dataset + '_loss.csv')