CN-122009928-A - AI-based elevator peak elevator traffic dynamic scheduling system

CN122009928ACN 122009928 ACN122009928 ACN 122009928ACN-122009928-A

Abstract

The invention relates to the technical field of elevator intelligent control, and particularly discloses an AI-based elevator peak boarding flow dynamic scheduling system. The system comprises an edge perception and feature extraction module, a cloud model collaborative optimization module, a dynamic scheduling decision module and a system resource self-adaptive allocation module, wherein high-precision passenger flow prediction and dynamic elevator dispatching are realized through a multi-level collaborative architecture, computing resources are dynamically adjusted according to the load of edge nodes, and passenger waiting time, system energy consumption and load balancing degree are optimized while real-time performance is ensured.

Inventors

CHEN ZHEN
TANG BO
WEI SHAOPENG
MA HAIWEN
ZHANG WEI
LI JINYUAN
ZHOU PENG
CHEN NING
LI YU

Assignees

中铁城建集团华东建设有限公司

Dates

Publication Date: 20260512
Application Date: 20260119

Claims (10)

1. An AI-based elevator peak boarding flow dynamic scheduling system, comprising: The edge perception and feature extraction module is arranged at edge computing nodes of the controller side of each elevator group of the building and is used for collecting elevator running state data, car inner load data, call request data of each floor and queuing people statistics information obtained through a vision sensor arranged in a hall, preprocessing and feature extraction are carried out on collected original data, and multidimensional feature vectors comprising time sequence features, space distribution features and passenger flow intensity features are generated; The cloud model collaborative optimization module is operated on the central server and is used for receiving the feature vectors and the historical scheduling result data uploaded by the plurality of edge nodes, constructing a passenger flow prediction model based on the deep neural network, continuously optimizing network parameters through periodic incremental learning, and transmitting the optimized model parameters to the edge nodes in a differential updating mode; The dynamic scheduling decision module is integrated at the edge node and is used for loading a lightweight prediction model issued by the cloud, predicting elevator taking demands in a future specific time window based on real-time feature vectors and generating a preliminary elevator taking scheme according to a preset scheduling objective function; and the system resource self-adaptive allocation module is used for monitoring the calculation load and the memory occupancy rate of the edge node in real time, and dynamically adjusting the number of the dimensions extracted by the features or the complexity level of the prediction model when the load is greater than a set threshold value.
2. The AI-based elevator peak boarding flow dynamic scheduling system of claim 1, wherein the feature extraction process performed by the edge awareness and feature extraction module comprises filtering, denoising and outlier rejection of raw sensor data; Extracting a passenger flow count sequence at intervals of 5 minutes from a time dimension; extracting call frequency distribution of each floor from space dimension, extracting change rate of current period relative to historical contemporaneous passenger flow from intensity dimension, normalizing the features and combining the features into feature vector with fixed dimension.
3. The AI-based elevator peak boarding flow dynamic scheduling system of claim 1, wherein the deep neural network adopted by the cloud model collaborative optimization module is a hybrid architecture of a long-term memory network and a self-attention mechanism; The deep neural network takes the multi-dimensional feature vector uploaded by the edge as input, captures the time dependency relationship through 3 layers of long-term memory network units, carries out weighted fusion on the importance of different features through a self-attention mechanism layer, and finally outputs the predicted value of the elevator taking demand of each floor in the period of 15 minutes in the future.
4. The AI-based elevator peak boarding flow dynamic scheduling system of claim 3, wherein the training process of the cloud model collaborative optimization module adopts a mean square error as a loss function, and performs a round of incremental learning by using the latest acquired data every 24 hours, and the updating amount of the model parameters is transmitted to the edge after being compressed.
5. The AI-based dynamic elevator rush hour traffic scheduling system of claim 1 wherein the scheduling objective function upon which the dynamic scheduling decision module is based is a multi-objective optimization function that considers both minimum average waiting time for passengers, minimum total energy consumption for the elevator system, and maximum 3 optimization objectives for each car load balancing.
6. The AI-based dynamic elevator rush hour traffic scheduling system of claim 5 wherein the dynamic scheduling decision module iteratively optimizes the preliminary dispatch protocol using an improved genetic algorithm.
7. The AI-based elevator peak boarding flow dynamic scheduling system of claim 1, wherein the system resource adaptive allocation module dynamically adjusts a system operation mode according to the central processor real-time occupancy rate of the edge node, when the central processor occupancy rate is greater than 80% for 10 seconds, the module automatically reduces the dimension of the feature vector from 128 dimension to 64 dimension and switches to a lightweight predictive model with lower computational complexity, and when the occupancy rate falls back below 60% and is maintained for 30 seconds, the system returns to a high-dimensional feature and complete model calculation mode.
8. The AI-based dynamic elevator rush hour traffic scheduling system of claim 1, further comprising a safety and fault tolerance control sub-module for continuously monitoring the operating status of the elevator group and the execution of the scheduling instructions; and immediately triggering an abnormal processing flow once detecting that a certain elevator does not respond to an elevator dispatching instruction for 2 continuous dispatching cycles or the actual load of the elevator car deviates from the predicted demand by more than 25%, putting the elevator into a manual mode and reassigning the calling task of the service floor.
9. The AI-based elevator peak boarding flow dynamic scheduling system of claim 1, wherein the system operates under a hierarchical time scale coordination framework, wherein cloud model optimization is performed in days, edge traffic prediction is performed in 5 minutes, dynamic scheduling decisions are generated in 15 seconds, and resource allocation and security monitoring are performed in real time in seconds.
10. The AI-based elevator peak boarding flow dynamic scheduling system of claim 2, wherein the process of extracting the change rate of the current period relative to the historical contemporaneous passenger flow from the intensity dimension comprises obtaining the current period passenger flow count value and the historical contemporaneous passenger flow mean value, calculating the difference between the current period passenger flow count value and the historical contemporaneous passenger flow mean value, dividing the difference by the historical contemporaneous passenger flow mean value and multiplying the difference by 100% to obtain the change rate.

Description

AI-based elevator peak elevator traffic dynamic scheduling system Technical Field The invention belongs to the technical field of intelligent elevator control, and particularly relates to an AI-based elevator peak elevator traffic dynamic scheduling system. Background In the field of intelligent architecture and vertical traffic automation, the scheduling efficiency of an elevator system directly affects the traffic experience and energy consumption of personnel in the building. Along with popularization of high-rise buildings and deep integration of internet of things, how to realize the intellectualization and self-adaption of elevator group control has become a key direction of industry development. The elevator dispatching system based on artificial intelligence aims to dynamically optimize elevator dispatching strategies by analyzing historical elevator taking data, real-time passenger flow information and building operation states so as to reduce passenger waiting time, balance system loads and improve overall operation efficiency. The prior art generally adopts a mode of deploying a complex prediction model at a cloud or a central server to realize scheduling decision. In the prior art, when the model is lightened and deployed on edge equipment for reducing delay, the prediction accuracy of the model is greatly reduced due to strict limitation of computing resources, and passenger flow fluctuation suddenly appearing in the peak period is difficult to accurately capture and respond to. The precision loss is particularly prominent in typical peak scenes such as office building business trips in the morning and evening, scheduling decision errors are easy to cause, part of elevators are overcrowded, other elevators are idle, and the problem of uneven resource allocation is solved, so that the service reliability and the user experience of the system are seriously reduced. There is therefore a need for an elevator traffic management scheme that can maintain high precision dynamic scheduling capabilities in a resource-constrained edge environment. Disclosure of Invention The invention aims to overcome the defects of model prediction accuracy reduction, slow peak passenger flow response and uneven system resource allocation caused by strict limitation of calculation resources in the edge deployment scene of an elevator dispatching system in the prior art. The invention aims to provide an AI-based elevator peak boarding flow dynamic scheduling system, which realizes high-precision passenger flow prediction and dynamic scheduling under the condition of limited resources of edge computing nodes by constructing a multi-level collaborative intelligent decision-making architecture. In order to achieve the above purpose, the invention provides an AI-based elevator peak boarding flow dynamic scheduling system, which comprises an edge perception and feature extraction module, a cloud model collaborative optimization module, a dynamic scheduling decision module and a system resource self-adaptive scheduling module. The edge sensing and feature extraction module is arranged at edge computing nodes of the controller side of each elevator group of the building and used for collecting elevator running state data, car load data, call request data of each floor and queuing people counting information obtained through visual sensors arranged in a hall. The module further performs preprocessing and feature extraction on the collected original data to generate a multidimensional feature vector comprising time sequence features, spatial distribution features and passenger flow intensity features. The cloud model collaborative optimization module operates on a central server, receives feature vectors and historical scheduling result data uploaded by a plurality of edge nodes, and builds a passenger flow prediction model based on a deep neural network. The model continuously optimizes network parameters through periodic incremental learning, and transmits the optimized model parameters to each edge node in a differential updating mode. The dynamic scheduling decision module is integrated at the edge node, loads a lightweight prediction model issued by the cloud, predicts elevator taking demands in a future specific time window based on real-time feature vectors, and generates a preliminary elevator taking scheme according to a preset scheduling objective function. The system resource self-adaptive allocation module monitors the calculation load and the memory occupancy rate of the edge node in real time, and dynamically adjusts the number of dimensions extracted by the features or the complexity level of the prediction model when the load is greater than a set threshold value so as to ensure the instantaneity of a scheduling decision. Further, the feature extraction process executed by the edge perception and feature extraction module specifically includes the following steps: firstly, filtering, denoising and outlier rejection