Update README.md

Modify the Requirements section.

README.md

@@ -3,7 +3,8 @@
 This is a TensorFlow implementation of Deep Neural Network in the following paper: 
 Licheng Qu, Wei Li, Wenjing Li, Dongfang Ma, and Yinhai Wang. [Daily long-term traffic flow forecasting based on a deep neural network](https://doi.org/10.1016/j.eswa.2018.12.031). Expert systems with applications, Volume 121, 1 May 2019, Pages 304-312.
 
-The future trend of traffic flow is closely related to historical data and its contextual factors. These additional contextual factors, such as collection time, day of the week, weather, season, etc., usually make the traffic flow exhibit various periodic and random fluctuations, and play a very important role in the prediction of traffic flow.
+The future trend of traffic flow is closely related to historical data and its contextual factors. These additional contextual factors, such as collection time, day of the week, weather, season, etc., usually make the traffic flow exhibit various periodic and random fluctuations, and play a very important role in the prediction of traffic flow. **This is the first time that only contextual factors have been used to predict traffic. 
+**This is the first time that only contextual factors have been used to predict traffic. Although this sounds mysterious and impossible, it has brought us surprising results.** 
 
 ## Optimization
 In this article, the optimal number of neurons and hidden layers is $[15, 18, 22, 9, 5]$, you can use the ***traffic_dnn_train_try.py*** program to find your own DNN structure layer by layer. This is a very simple search routine, just trying to find a suitable next layer within a limited range based on the initially defined hidden layer.
@@ -23,7 +24,7 @@ After the training, you can predict or evaluate the model by program ***traffic_
 Because the Table 1 of the paper only only evaluate the traffic between 6:00 and 22:00. 
 
 ## Requirements
-These programs are developed based on Temsorflow 1.2. It is recommended to install Anaconda and use Pycharm to evaluate it. If you want to use the latest tensorflow 2.x, please use its compatibility mode.
+These programs are developed based on TensorFlow 1.2. Numpy, pandas and Matplotlib are also essential scientific computing libraries. It is strongly recommended to use Anaconda to manage all software packages and use Pycharm IDE for code evaluation. If you have installed the latest TensorFlow 2.x, please use its compatibility mode.
 ```
  #import tensorflow as tf
  import tensorflow.compat.v1 as tf