Multi-mode AI Regulation Business Engine-多模式AI调控业务引擎

Three Business

Comprehensive Infrastructure

Rail Transit Network

Major Transportation Corridors

Smart Neighborhoods

New Infrastructure Engineering

Digital and Intelligent Governance

Comprehensive Management and Control

Vehicle-Road-Cloud

Low-altitude Control and Management

Power Charging, Storage and Discharging

New - Quality Operations

Asset Activation

Low-altitude Operation

Lifeline Engineering

Operation Assurance

Digital Base

Multi-mode AI Regulation Business Engine

Evolutionary Computing Platform

Digital Twin Foundation

Multi-mode AI Regulation Business Engine

The current development of transportation is evolving towards cross-domain integration, system resilience, and autonomy. However, it still faces challenges such as "unclear interactive feedback mechanisms of complex systems", "inaccurate traffic prediction under abnormal conditions", and "insufficient intelligence level of transportation systems". In response to the pain points in the transportation industry, Shenzhen Urban Transport Planning Center Co., Ltd. has built a multi-mode AI regulation business engine based on digital twins, driven by data + mechanism + knowledge, with large models + small models as the core, and embodied intelligent agents as the executors.

Overall Solution

A multi-mode AI regulation business engine built on the TransPaaS platform, integrating semantic knowledge bases, large model services, online distillation and evolution capabilities, and multi-source traffic operation data, to create a full-chain capability system covering "knowledge Q&A - travel prediction - operation evaluation - scheme generation", supporting the transformation of urban-level intelligent traffic governance and the upgrading of integrated services.

Intelligent Q&A

It provides intelligent semantic retrieval and knowledge-based Q&A services on traffic policies, rules, and business processes for the public, municipal staff, and enterprise users. It enables complex intent parsing and multi-turn dialogue interaction, establishing an efficient, convenient, and sustainably updated intelligent knowledge service channel to comprehensively enhance the information service capability of traffic governance.

Cross-Domain Integrated Deduction

For the traffic operation status across multiple scenarios such as urban roads, rail transit, and expressways, a multi-mode prediction model integrating factors like travel behavior, meteorological environment, holiday periods, and hotspots is constructed. It supports traffic trend prediction and operation analysis at multiple scales and time granularities, providing a priori support for management strategies.

Transportation-Energy Integration Planning and Evaluation

For the planning and construction tasks of transportation facilities such as new rail lines, transportation hubs, and industrial parks, an evaluation and deduction system integrating multi-dimensional data is built. The system supports evolutionary simulation and comparison of multiple schemes, and can provide a basis for comprehensive decision-making under complex planning conditions, improving the efficiency of urban spatial resource allocation.

Large-scale Hub Capacity Optimization

Based on real-time perception and prediction results, for urban-level passenger-intensive hub scenarios, an intelligent regulation system integrating demand identification, capacity generation, and task scheduling is constructed. It supports functions such as personalized reservations, flexible scheduling, and automatic push, serving the efficient response and supply-demand matching of urban traffic operations.

Implemented Cases

Intelligent Q&A

Traffic Large-Model Multi-Mode Prediction System

Transportation-Energy Integration Planning and Evaluation System

Large-scale Hub Capacity Optimization Platform

To address citizens' daily needs such as transportation policy consultation, opinion collection, and business handling, a "Traffic Intelligent Q&A Platform" has been developed and deployed. The platform integrates the large-model capabilities of DeepSeek, supports dual-mode input of voice and text, automatically identifies the business intentions behind users' questions, and accurately matches government affairs knowledge, process guidelines, and data analysis results. It has now been put into external service, enabling citizens to get "instant answers" for their affairs. Compared with traditional transportation service hotlines, it has improved response efficiency and enhancing user satisfaction.

The large-model prediction system based on multi-factor integration comprehensively considers factors such as holiday passenger flow, weather changes, events in popular scenic spots, and population mobility, supporting multi-type traffic flow prediction including normal conditions, few-samples, and zero-samples. The technology has been applied in the "Guangdong-Hong Kong-Macau Expressway Intelligent Traffic Control Project" and the "Shenzhen NOCC Phase II Construction". It is widely used in predicting road network operation trends during holidays and formulating multi-level scheduling strategies, providing citizens with more accurate travel suggestions and significantly alleviating node congestion.

Centering around Shenzhen's urban renewal and energy strategic deployment, a multi-factor integrated deduction and evaluation engine has been constructed. This system has served the optimization design of rail networks by integrating POI distribution, land use, network structure, and population behavior data to predict the traffic flow evolution trend after the opening of new lines. In the field of transportation energy, the platform supports regional energy demand evolution modeling based on data such as activity hotspots, vehicle and pedestrian distribution, and meteorological conditions. It has assisted Shenzhen in completing a special plan for transportation energy, improving the efficiency of energy supply-dema

In the practice of passenger flow management during peak hours at Shenzhen North Railway Station, the automatic generation and application of intelligent scheduling schemes have been realized. Based on traffic prediction results, the system links with APP platforms to issue early warnings of large passenger flows to taxi drivers, supporting drivers in dispatching capacity according to reservation information. Pilot data shows that after the application of this function, the empty driving rate of taxis has significantly decreased, and the average order acceptance efficiency of drivers has increased by about 1.8 times. This scheme can be promoted as a model in multiple key hubs across the city, thereby enhancing the flexible control capability and service quality of urban transportation.

Our Advantages

Analysis capability for few-sample/zero-sample events based on traffic scenario-oriented thought chains and transfer learning

Self-learning and self-evolving capabilities based on reinforcement learning + embodied intelligence

Traffic operation mechanism learning capability centered on cross-domain knowledge graphs and causal inference