Development of an AI-based urban planning system
Urban planning traditionally relies on outdated data and expert judgment. AI brings objective spatial data analytics and predictive power of urban planning decisions to this process.
Urban environment analysis
Quality of the urban environment (KGS Index):
The Russian Ministry of Construction has introduced an Urban Environment Quality Index (UEQI) consisting of 36 indicators. AI automates the collection of most of the data:
- Street View analysis: Google Street View / Yandex Panoramas → Computer Vision assessment of landscaping: - Presence of trees, sidewalk surfaces, facade condition, lighting - ResNet50 trained on assessments by urban planning experts → automatic scoring
import torch
import torchvision.models as models
import torchvision.transforms as transforms
from PIL import Image
import requests
class StreetViewQualityAnalyzer:
"""Оценка качества городской среды по снимкам street view"""
QUALITY_ASPECTS = ['greenery', 'walkability', 'lighting',
'building_condition', 'cleanliness', 'safety_perception']
def __init__(self, model_path):
self.model = models.resnet50(pretrained=False)
self.model.fc = torch.nn.Linear(2048, len(self.QUALITY_ASPECTS))
self.model.load_state_dict(torch.load(model_path))
self.model.eval()
self.transform = transforms.Compose([
transforms.Resize(256), transforms.CenterCrop(224),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
])
def score_location(self, lat, lon, yandex_api_key):
"""Оценить качество среды в точке по панорамным снимкам"""
# Получить panorama ID от Яндекс.Карт
url = f"https://api.maps.yandex.ru/1.x/?apikey={yandex_api_key}&ll={lon},{lat}&type=panorama"
panorama_img = self._fetch_panorama(url)
if panorama_img is None:
return None
x = self.transform(panorama_img).unsqueeze(0)
with torch.no_grad():
scores = torch.sigmoid(self.model(x))[0]
return dict(zip(self.QUALITY_ASPECTS, scores.tolist()))
Pedestrian Comfort Index:
Thermal stress, noise, air pollution + spatial proximity of active facades → integrated pedestrian comfort index. Identification of "dead" zones for priority improvement.
Modeling of building density
Floor Area Ratio (FAR) optimization:
When designing a new neighborhood: what is the optimal building density?
Agent-based simulation (Mesa + NetworkX): - Agents: residents, cars, pedestrians - Environment: street network, transport, POI - Simulation of life under different development scenarios → assessment of infrastructure load
Solar Access Analysis:
3D model of the building + solar calculation → insolation analysis: - Violation of SanPiN 2.2.1/2.1.1.1076 for insolation of apartments - Prism simulator (LadyBug for Grasshopper/Rhino) + ML scoring - Optimization of building height and offsets to comply with standards
Transport planning
Trip Generation Modeling:
How many trips does a new development generate? - Regression model on ITE (Institute of Transportation Engineers) trip data - Inputs: facility type, area, location (central/peripheral), public transport accessibility - Outputs: peak hour car trips → required capacity
Network Analysis:
- Connectivity analysis: how will travel time change with a new road/route? - Betweenness centrality of the road graph → critical links - Vulnerability: what happens if X nodes are disconnected (resilience to accidents)
Decision making
Multi-Criteria Decision Analysis (MCDA):
Selecting a location for a new facility (school, park, transport hub): - Criteria: accessibility for the target group, land cost, load on infrastructure - Criteria weights: expert input + AHP (Analytic Hierarchy Process) - Priority map: where is the best place to locate the facility
Generative Urban Design:
Setting parameters (PLU (Plan Local d'Urbanisme), standards, budget) → AI generates options for developing the block: - Diffusion models for generating planning schemes - Evaluator: compliance with standards + environmental quality score - Filtering: only options that comply with standards are considered by the planner
Development time: 5–9 months for an Urban Planning AI system with spatial analysis, street view scoring, and scenario modeling.