To predict the poi distribution of Xi 'an in the future based on deep learning network

1. Summary:

   With the development of cities, commercial buildings play an important role, and the selection of commercial building land has a significant impact on commercial buildings.

   However, finding suitable commercial land is a significant challenge for designers. In traditional design, selecting land requires extensive investigation in the early stages, so we hope to use GAN to identify potential commercial circles. During the experiment, our main question was whether GAN could assess potential business circles in regions without established business circles by learning information about existing business circles.

   We trained GAN by importing CSV files of Weinan POI data and classifying the Weinan commercial area. After the training is completed, input the map of Weinan City into the trained model for exploration, and finally output the commercial areas that may appear in the Weinan urban area and surrounding areas. This study verified the feasibility of using GAN based commercial land selection through empirical research, providing useful references for urban planning and renewal.

   During the experiment, our main question was whether GAN could assess potential business circles in regions without established business circles by learning information about existing business circles. We train GAN by importing CSV files of Xi'an POI data and classifying the commercial district of Xi'an. After the training is completed, input the map of Weinan City into the trained model for exploration, and finally output the commercial areas that may appear in the Weinan urban area and surrounding areas.

   This study verified the feasibility of using GAN based commercial land selection through empirical research, providing useful references for urban planning and renewal.

   Keywords: GAN machine learning urban planning

   Research background:

   The selection of commercial land is crucial for commercial building design, and a reasonable choice can lay a solid foundation for the construction and operation of commercial buildings. Firstly, the location of commercial land directly affects the passenger flow of commercial buildings. Secondly, the surrounding environment has a significant impact on the image and attractiveness of commercial buildings. In addition, the selection of commercial land also needs to consider market demand and development trends. Last but not least, the size of commercial land determines the scale and functionality of the building.

   Commercial architectural design is an important driving force for urban development. It can promote comprehensive urban development by shaping the urban environment, promoting community development, inheriting urban culture, and coordinating with urban planning. Commercial buildings are an important component of urban landscape. Excellent commercial architectural design can provide a unique visual experience for the city, enrich the diversity of urban space, and enhance the attractiveness of the city.

   Our study takes Weinan City as an example and selects the layout trends of five commercial types to provide some ideas for the future urban commercial layout overview.

   We use Generative Adversarial Networks (GANs) to predict potential business circles in regions without established business circles. We have established a model to generate a POI scatter map of the city, and trained it with GAN to obtain the possibility of similar commercial layouts.

   Finally, we need an effective method to validate our results. We validate the effectiveness of the model by comparing the predicted results with actual commercial areas. We are also considering using some additional data, such as sales data, pedestrian flow, etc., to further validate the model's predictions.

   Project logic:

   Firstly, we extract the POI data CSV format of Weinan City from the Gaode map, and annotate and select the type code. 

2. 

   Then encode the POI data from the CSV file, normalize the longitude and latitude, divide the map area based on the normalized data, and draw scatter plots for each area. Classify the Weinan commercial area and save it as an image file.

   

   Load the image dataset of the map area. In addition, image data is preprocessed by resizing, converting to tensors, and normalizing.

   We used Generative Adversarial Networks (GANs) to predict potential business districts in areas without established business districts, and selected a generator to generate new business district image

   Code content：代码内容：

论Ai生成城市商业布置方位可能性 ——以渭南市区为例

摘要:

随着城市的发展，商业建筑发挥着重要的作用，商业建筑用地的选择对商业建筑有着重要的影响。

然而，寻找合适的商业用地对设计师来说是一个重大挑战。在传统设计中，选择土地需要在早期进行广泛的调查，因此我们希望使用GAN来识别潜在的商业圈。在实验过程中，我们的主要问题是GAN是否可以通过学习现有商业圈的信息来完成对没有建立商业圈的地区潜在商业圈的判断。

我们通过导入渭南POI数据的CSV文件并对渭南商业区进行分类来训练GAN。训练完成后，将渭南市地图输入到训练的模型中进行探索，最终输出渭南城区及周边可能出现的商业区域。本研究通过实证研究验证了基于GAN的商业用地选择的可行性，为城市规划和更新提供了有益的参考。

在实验过程中，我们的主要问题是GAN是否可以通过学习现有商业圈的信息来完成对没有建立商业圈的地区潜在商业圈的判断。我们通过导入西安POI数据的CSV文件并对西安商业区进行分类来训练GAN。训练完成后，将渭南市地图输入到训练的模型中进行探索，最终输出渭南城区及周边可能出现的商业区域。

 

本研究通过实证研究验证了基于GAN的商业用地选择的可行性，为城市规划和更新提供了有益的参考。

关键词：GAN机器学习城市规划

研究背景：

商业用地的选择对商业建筑设计至关重要，合理的选择可以为商业建筑的建设和运营奠定坚实的基础。首先，商业用地的区位直接影响商业建筑的客流量。其次，周边环境对商业建筑的形象和吸引力有显著影响。此外，商业用地的选择还需要考虑市场需求和发展趋势。最后但同样重要的是，商业用地的大小决定了建筑的规模和功能。

商业建筑设计是城市发展的重要推动力。它可以通过塑造城市环境、促进社区发展、传承城市文化、与城市规划相协调来促进城市的全面发展。商业建筑是城市景观的重要组成部分。优秀的商业建筑设计可以为城市提供独特的视觉体验，丰富城市空间的多样性，增强城市的吸引力。

我们此次的研究以渭南市为例选取其5种商业类型的布置趋向来对未来城市商业布置概况提供一些思路。

我们使用生成对抗性网络（GAN）来预测没有建立商业圈的地区的潜在商业圈。我们建立了模型来实现生成城市的poi散点图，通过gan训练使其得到类似的商业布置的可能性。

最后，我们需要一种有效的方法来验证我们的结果。我们通过将预测结果与实际商业区域进行比较来验证该模型的有效性。我们还考虑使用一些额外的数据，如销售数据、行人流量等，以进一步验证模型的预测。

项目逻辑：

• 首先我们从高德地图中提取渭南市的poi数据csv格式，并标注并选取typecode的类型。

• 然后编码从CSV文件中读取POI数据，对经纬度进行归一化，根据归一化后的数据划分地图区域，并绘制每个区域的散点图，对渭南商业区进行分类并保存为图像文件。

• 加载地图区域的图像数据集。此外，通过调整大小、转换为张量和归一化来预处理图像数据。
• 使用生成对抗性网络（GAN）来预测没有建立商业圈的地区的潜在商业圈，我们选定了一个生成器用于生成新的商业区图像
  • After the success of the project, we will consider validating the effectiveness of the model by comparing the predicted results with actual commercial areas.
    
  • If we want to further derive the model, we will also consider using some additional data, such as sales data, pedestrian flow, etc., to further validate the model's predictions.
  • Experimental steps:
  • Encode the POI data from a CSV file, normalize the longitude and latitude, divide the map area based on the normalized data, and draw scatter plots for each area. Classify the Xi'an commercial area and save it as an image file. Load the image dataset of the map area. In addition, image data is preprocessed by resizing, converting to tensors, and normalizing.
  • 该代码定义了两个模型：一个生成器用于生成新的商业区图像，另一个鉴别器用于确定输入图像是真实的还是生成的。此代码为生成器和鉴别器定义了一个 Adam 优化器，并定义了一个二进制交叉熵损失函数来测量生成器和鉴别器的性能。因此，生成器可以生成类似于商业布局的散点图。
  • 该代码使用对抗网络在循环中交替训练生成器和鉴别器。在每个训练周期中，代码执行前向传播、计算损失、执行反向传播并优化参数。
  • 在该项目成功后，我们考虑通过将预测结果与实际商业区域进行比较来验证该模型的有效性。
  • 如要进一步推导该模型我们还考虑使用一些额外的数据，如销售数据、行人流量等，以进一步验证模型的预测。
  实验步骤： 编码从CSV文件中读取POI数据，对经纬度进行归一化，根据归一化后的数据划分地图区域，并绘制每个区域的散点图，对西安商业区进行分类并保存为图像文件。加载地图区域的图像数据集。此外，通过调整大小、转换为张量和归一化来预处理图像数据。 代码定义了两个模型：一个生成器用于生成新的商业区图像，另一个鉴别器用于确定输入图像是真实的还是生成的。该代码为生成器和鉴别器定义了Adam优化器，并定义了一个二进制交叉熵损失函数来测量生成器和鉴别器的性能。从而生成器可以生成类似与商业布置的散点图。 代码使用对抗性网络训练循环交替训练生成器和鉴别器。在每个训练周期中，代码执行前向传播、计算损失、执行反向传播并优化参数。