tensorflow实现非线性回归实例

import tensorflow as tf
import numpy as np
import os
import matplotlib.pyplot as plt
os.environ['TF_CPP_MIN_LOG_LEVEL']='2'

# R25=10k B25/50=3470 NTC热敏电阻特性数据表

# 加载样本数据 格式：电阻值(K)   温度值(℃)
dat = np.loadtxt('trainData.txt')


# 数据归一化处理
R = dat[:,0]
T = dat[:,1]

R_K = R.max()-R.min()
R_B = R.min()

T_K = T.max()-T.min()
T_B = T.min()

R = (R - R_B)/R_K
T = (T - T_B)/T_K

R = R.reshape(141,1)
T = T.reshape(141,1)

X = tf.placeholder(tf.float32, shape = [None, 1])
Y = tf.placeholder(tf.float32, shape = [None, 1])

# 定义层
def add_layer(input,in_size,out_size,activation_fun):
    """

    :param input: 输入数据
    :param in_size: 输入矩阵列数
    :param out_size: 输出矩阵列数
    :param activation_fun:激活函数
    :return:输出矩阵
    """

    weights = tf.Variable(tf.random_normal([in_size, out_size]))
    bias = tf.Variable(tf.zeros([1, out_size]))
    z_i = tf.matmul(input, weights) + bias

    return activation_fun(z_i)


# 正向传播

#  添加隐藏层 该层10个神经元
out_h = add_layer(X,1,10,tf.nn.sigmoid)

# 输出层
out = add_layer(out_h,10,1,tf.nn.sigmoid)

# 定义损失函数
loss = tf.reduce_mean(tf.reduce_sum(tf.square(out - Y), reduction_indices=[1]))

# 学习率为
learning_rate = 20.2

# 梯度下降优化器，让损失最小化
train_step = tf.train.GradientDescentOptimizer(learning_rate).minimize(loss)

# 初始化tensor flow中的变量
init = tf.global_variables_initializer()

with tf.Session() as sess:
    sess.run(init)
    for i in range(10000):
        sess.run(train_step, feed_dict={X: R, Y: T})
        if i % 100== 0:
            # 打印损失值
            e = sess.run(loss, feed_dict={X: R, Y: T})
            print(e)

    # 预测值
    y = sess.run(out, feed_dict={X: R, Y: T})
    y = y*T_K+T_B

    dat_R = dat[:,0]
    dat_T = dat[:,1]

    plt.rcParams['font.sans-serif']=['SimHei']
    plt.rcParams['axes.unicode_minus'] = False
    plt.xlabel("电阻")
    plt.ylabel("温度")
    plt.plot(dat_R ,dat_T,'r',label='R-T特性曲线')
    plt.plot(dat_R,y,label='拟合曲线')
    plt.legend()
    plt.grid()
    plt.show()

运行结果：