PSO-BP 回归预测 python语言实现

import math; import random; from matplotlib import pyplot as plt from tensorflow import keras from keras.optimizers import Adam from core.utils import Timer from keras.models import Sequential # from keras.models import Sequential from keras.layers.core import Dense import numpy as np import matplotlib.animation as animation from PIL import Image # from keras.optimizers import Adam from keras.callbacks import ModelCheckpoint from keras.utils import np_utils import os import pandas as pd # from tensorflow.examples.tutorials.mnist import * from keras.models import * from keras.layers import * from keras import callbacks # import imageio # from PyQt5 import QtWidgets import time import os import psutil def show_info(start): pid = os.getpid() #模块名比较容易理解：获得当前进程的pid p = psutil.Process(pid) #根据pid找到进程，进而找到占用的内存值 info = p.memory_full_info() memory = info.uss/1024 return memory def count_time(func): def int_time(*args, **kwargs): start_time = time.time()#这个是程序开始时间 func() over_time = time.time()#这个是程序结束时间 total_time = over_time - start_time # print('程序共计%s秒' % total_time) return int_time def count_info(func): def float_info(*args, **kwargs): pid = os.getpid() # 模块名比较容易理解：获得当前进程的pid p = psutil.Process(pid) # 根据pid找到进程，进而找到占用的内存值 info_start = p.memory_full_info().uss/1024 func() info_end=p.memory_full_info().uss/1024 # print('此程序运行占内存'+str(info_end-info_start)+'kB') return float_info # def create_gif(path, gif_name, duration): # frames = [] # image_list = [] # for i in range(0,1999): # image_list.append(path+"/%d.jpg" % i) # for image_name in image_list: # frames.append(imageio.imread(image_name)) # imageio.mimsave(gif_name, frames, 'GIF', duration=duration) # return inputfile = 'input.xlsx' # excel输入 outputfile = 'output.xls' # excel输出 modelfile = 'modelweight.model' os.chdir( "D:/nn_bp"); # training_set = pd.read_csv("Social_Network_Ads.csv"); training_set1 = pd.read_csv("D:/nn_predict/traindata.csv") # pandas以DataFrame的格式读入excel表 feature = ['ac', 'cnl', 'gr', 'rs', 'den'] # 影响因素四个 label = ['so2'] # 标签一个，即需要进行预测的值 depth = ['depth'] training_set = training_set1.loc[range(0, 54)].copy() # 标明excel表从第0行到58行是训练集 # print(training_set[label]) # 2 数据预处理和标注 data_mean = training_set.mean() data_std = training_set.std() data_train = (training_set - data_mean) / data_std # 数据标准化 y_data_train = (training_set) / training_set.max() k = training_set[label].max().values def getfit(model, x): # x = ((data_train[feature] - data_mean[feature]) / data_std[feature]).as_matrix() # y = (model.predict(x)) * data_std[label].values + data_mean[label].values y=(model.predict(x))*k return y def init(): fpath = "D:/nn_bp/image/0.jpg"; img = Image.open(fpath); plt.axis('off')#  关掉坐标轴为 off return plt.imshow(img); def update(i): fpath = "D:/nn_bp/image" + str(i) + ".jpg"; img = Image.open(fpath); plt.axis('off') #  关掉坐标轴为 off return plt.imshow(img); def bp_main(x1,x2): # print(k) start = show_info('开始') x_train = data_train[feature].iloc[:, :].values # 特征数据 y_train = y_data_train[label].iloc[:, :].values # 标签数据 depth_test = training_set[depth] y_plot = training_set[label] # print(x_train) # print(y_train) #########预测数据x predict_set = pd.read_csv("D:/nn_predict/traindata+juedui.csv") feature1 = ['a', 'b', 'cc', 'd', 'e'] predict_set_feature = predict_set.loc[range(0, 57)].copy() predict_mean = predict_set_feature.mean() predict_std = predict_set_feature.std() predict_data = (predict_set_feature - predict_mean) / predict_set predict_data_x = predict_data[feature1].iloc[:, :].values filepath = "D:/nn_bp/model/weights-improvement-{epoch:00d}.hdf5"; callbacks_list = [ # 监视变量monitor在patience个epoch间差值都小于阈值min_delta时停止训练， # mode即该变量的类型，loss是min 越小越好，max越大越好 例如 精度 callbacks.EarlyStopping( monitor='val_loss', min_delta=0.000001, patience=2, verbose=1, mode='min' ), # 保存训练过程中的最优模型，filepath保存路径 # monitor监视变量，保存训练过程中该值最优的模型， # save_weights_only False相当于model.save()保存整个模型结构和权重 True 只保存权重 # save_best_only 是否覆盖当前最好的模型 callbacks.ModelCheckpoint( filepath=filepath, verbose=1, monitor='val_loss', save_weights_only=False, mode='auto', save_best_only=False ), # 将训练过程的精度loss等写入文件中，append 当文件存在时 是否在尾部添加或是覆盖 callbacks.CSVLogger( filename='loss/train.csv', append=True ), # 降低学习率，当监视变量在patience个epoch中都没有改善（根据min_delta阈值判断）时就降低学习率， # lr = lr*factor，此处即减10倍，min_lr 学习率下限 callbacks.ReduceLROnPlateau( monitor='val_loss', factor=0.1, # lr*factor patience=10, mode='min', min_delta=1e-4, min_lr=0 ) ] low = 0; up = 2 * math.pi; # callback # filepath = "D:/user/测井/中海油/nn_bp/model/weights-improvement-{epoch:00d}.hdf5"; # checkpoint = ModelCheckpoint(filepath, verbose=1, save_best_only=False, mode='auto'); # callbacks_list = [checkpoint]; # #  建立顺序神经网络层次模型 # model = Sequential() # 层次模型 # model.add(Dense(64, input_dim=5, init='uniform',activation='sigmoid')) # 输入层，Dense表示BP层###目的：找到创新思路 问题：文献解决问题 到底有无创新性 思考：为什么作者找到创新方法 创新方法有无弊端 有无新的方法 # model.add(Dropout(0.5)) # # model.add(Dense(32, init='uniform', activation='sigmoid')) # model.add(Dropout(0.2)) # # model.add(Dense(1, activation='tanh')) # 输出层 timer = Timer() timer.start() adam_lr = x1 adam_beta_1 = x2 # model.compile(loss='mean_squared_error', optimizer=Adam(lr=adam_lr, beta_1=adam_beta_1), metrics=['accuracy']) # 编译模型loss='mean_squared_error' model=Sequential() model.add(Dense(input_dim=5, units=32, activation="relu"))######特征数 # model.add(Dropout(0.5)) model.add(Dense(units=16, activation="relu")) # model.add(Dropout(0.5)) # model.add(Dense(units = 16, kernel_initializer = "uniform", activation="relu")) # # # model.add(Dropout(0.5)) # model.add(Dense(units = 8, kernel_initializer = "uniform", activation="relu")) # model.add(Dense(units=8, kernel_initializer="uniform", activation="relu")) # model.add(Dropout(0.5)) model.add(Dense(units=1, activation="tanh"))#sigmoid model.compile(optimizer=Adam(lr=adam_lr, beta_1=adam_beta_1), loss="mean_squared_error", metrics=["accuracy"]) his = model.fit(x_train, y_train, epochs=2000, batch_size=5, callbacks=callbacks_list,verbose=1,) # 训练模型1000