Abstract : A global control algorithm for a twin elevator group was designed using particle swarm optimization (PSO). The system automatically calculates the number of passengers waiting for the elevator based on floor video images. It generates a data stream containing passenger waiting and riding information and autonomously selects either an adaptive mutated PSO algorithm or a hybrid PSO algorithm based on different traffic modes. The system completed the design of waiting data, algorithm selection and operation, and the overall system flow. Simulation results compared the data with the multi-objective optimization control indicators of the elevator group. The results show that the two improved PSO algorithms achieve better results for the control objectives, verifying the effectiveness and applicability of the algorithms.
Keywords: twin elevators; particle swarm optimization algorithm; control system; elevator group
1. Introduction
With the rapid economic development and the improvement of urban modernization, high-rise and super high-rise landmark buildings, as symbols of economic strength and technological advancement, have emerged, thus placing higher demands on the service quality of vertical transportation in these buildings. Statistics [1-2] show that elevators in high-rise and super high-rise buildings consume approximately 6%-15% of the building's total energy consumption, a problem that cannot be ignored. Elevator shafts occupy more than 31% of the building's core area, significantly reducing the actual usable area and lowering the building's value. For example, the Burj Khalifa in Dubai, with its heavy passenger flow, houses 56 ultra-high-speed elevators, and the cost of its elevator shafts is as high as tens of millions of yuan per square meter.
To overcome the bottleneck of elevators restricting urban development towards the sky, the Twin Elevator, considered the "ultimate dream" in the elevator industry, was developed. ThyssenKrupp Group manufactured the world's first Twin Elevator system in 2002, which has been successfully applied in projects such as the University of Stuttgart in Germany, the Federation Tower in Russia, the Valencia Marine Center in Spain, the Trumpf Technology Center in Seoul, the Main Triangel Building in Frankfurt, and the Xinghai Holiday Hotel in Dalian. Due to its structural characteristics, it has a higher carrying capacity and more complex operating rules than traditional elevators. How to achieve efficient scheduling of Twin Elevator systems, optimize the control of Twin Elevator groups, improve the passenger experience and operating efficiency of high-speed and ultra-high-speed elevators, and reduce energy consumption has become a key research focus and hot topic in the elevator industry both domestically and internationally. Domestic research on Twin Elevators is extremely limited (currently only 18 domestic papers of various types can be found). This paper applies particle swarm intelligent control algorithms to the research of Twin Elevator group control systems to reduce the technological gap with foreign countries, provide a theoretical basis for the engineering scheduling of Twin Elevator systems in my country, and promote the development of elevator group control technology in my country.
2. Particle Swarm Optimization Algorithm
Inspired by bird predation behavior, Kennedy and Eberhart[3] first proposed the Particle Swarm Optimization (PSO) algorithm based on the single bird flight trajectory with the shortest distance to food and flock cooperation. The PSO algorithm abstracts birds into "particles" with only position and velocity in the search space. It initializes a group of particles representing each possible solution in the solvable space and uses position points (which can be understood as the particles themselves), velocity that determines flight direction and distance, and fitness design function values to express their characteristics. In each iteration, the particles track the individual extreme value of the best position flown by in the current generation and the global extreme value of the best position experienced by the population in the current generation to update their own position and velocity, thereby gradually approaching the optimal position and finding the optimal solution to the optimization problem. The PSO algorithm avoids complex genetic operations compared with the traditional population evolution algorithm through the position and velocity model. It can dynamically track the current search situation and adjust the search strategy in time. The algorithm has fewer parameters to adjust and a simple structure, so it can find the global optimal solution faster.
Particle swarm optimization (PSO) algorithms have advantages such as simple structure, fast convergence speed, few parameter settings, and strong adaptability. Currently, PSO algorithms are widely used in signal processing, multi-objective optimization, video image processing, and intelligent robotics. However, few experts and scholars both domestically and internationally have utilized PSO algorithms for the optimized scheduling of elevator group control systems.
3. Design of a global control algorithm for twin elevators based on particle swarm optimization
The twin elevator group control system based on particle swarm optimization algorithm is mainly divided into a passenger flow data generation module, a control algorithm module, an operation control module, and an operation result statistics module. Through the coordination and real-time communication of these sub-modules, the reliable operation and efficient control of the twin elevator group control system are finally achieved.
3.1 Operation Flow of Particle Swarm Control Algorithm
In the optimization of elevator group control systems using the particle swarm optimization algorithm, each particle represents a dispatching scheme. Each particle's code is divided into upward and downward parts. The particle's dimension depends on the number of call signals currently being processed, but can also be set to a fixed value. The algorithm first randomly assigns an elevator to each call task based on real-time passenger flow data, thus randomly generating an initial population of a certain size. Based on the analysis of the group control system's performance evaluation indicators, a weighted coefficient method or expert scoring method is used to linearly combine the evaluation functions of each elevator's performance indicators to obtain the fitness function of the particle swarm optimization algorithm. The fitness value of each particle in the population, i.e., each dispatching scheme, is calculated (the value represents the particle's quality); the higher the fitness value, the better the dispatching scheme corresponding to the particle. After selecting individual and global extreme values and updating the particles, the next generation of particle swarm is generated. The algorithm calculates until the iteration termination condition is met, and finally outputs the optimal elevator dispatching scheme. Otherwise, the inertia weight is reset, the initial particle swarm is updated, and the cycle continues until the average fitness value of the particle swarm reaches the set target value or the number of generations reaches the maximum value. Finally, the global optimization control of the twin elevator swarm is achieved.
3.2 Generation of Elevator Waiting Passenger Flow Data
A video image-based elevator waiting number statistics module is used to determine the number of people waiting for elevators on each floor. It is assumed that the pattern of passengers arriving at the lobby and waiting for elevators follows a Poisson distribution. By setting parameters such as passenger arrival rate and elevator call probability per unit time, the passenger flow generation function is substituted into the initial density vector and Monte Carlo sampling to generate real-time elevator passenger flow data including passenger arrival time, elevator call signal floor, and target floor (the elevator passenger flow data generation process is shown in Figure 1).
Figure 1. Generation of passenger flow data for elevator waiting areas.
3.3 Two Improved Particle Swarm Optimization Algorithms
Traditional particle swarm optimization (PSO) algorithms, as the number of system iterations increases, lead to a convergence and concentration of the particle population, while the position vectors of each particle become increasingly similar. This can result in the particle becoming trapped near local optima. Therefore, to maintain diversity and obtain an optimized dispatching scheme suitable for the operating rules of twin elevators, thereby improving the control effect of the twin elevator group control system, two improved PSO algorithms can be used: adaptive mutated PSO and hybrid PSO. The hybrid PSO algorithm is used when the traffic is in both directions or when the passenger flow density exceeds a large set value in the inter-floor mode; the adaptive mutated PSO algorithm is used when the passenger flow in the inter-floor mode is small and does not exceed the expert set value.
Among them [4-5], the adaptive mutation particle swarm optimization algorithm reinitializes some particles in the particle swarm with a certain probability after each sub-particle update, which can increase the solvable search space of the particles. The hybrid particle swarm optimization algorithm combines the genetic intelligence algorithm with the particle swarm optimization algorithm, introducing the concepts of crossover and mutation operators of the genetic algorithm. In the crossover operation, particles are first paired one-to-one, and then two crossover positions are randomly generated. The elements at the crossover positions of the two paired particles are swapped to obtain two new particles. In order to retain excellent sub-particles, particle updates are only performed if the fitness value of the new (sub)particle generated by crossover is greater than that of the original source (parent) particle; otherwise, the source (parent) particle is retained. The mutation operation adopts a two-way interactive mutation method. First, two mutation positions are randomly selected on any particle, and then the corresponding elements of the two mutation positions are swapped to complete the mutation operation. Similar to the crossover operation, only the new (sub)particles with larger fitness values after mutation are retained for updating (as shown in Figure 2). In order to improve the global search capability of the intelligent algorithm, the mutation probability should not be too small.
Figure 2. Flowcharts of the two improved particle algorithms
3.4 Global Design of Twin Elevator Group Control System
The operation control module is responsible for simulating elevator operation, including the acceleration and deceleration of the traction machine and door operator, door opening and closing, constant speed operation, and stopping functions. It can also monitor the car's position and operating status in real time. The operation result statistics module is responsible for recording the passenger's entry and exit times, the number of passengers entering and exiting the elevator, etc., during each stop of the car at the lobby. It can also calculate the passenger's waiting time and riding time, and count the number of stops for each elevator stop, providing data for the systematic evaluation of the performance of the twin elevator group control system. The overall flowchart of the twin elevator group control system is shown in Figure 3.
Figure 3. Flowchart of the overall design of the twin elevator group control system
4. Design of an automatic elevator occupancy counting device based on video images
While crowd control systems provide services based on passenger flow data, they currently cannot automatically generate such data, making manual statistics labor-intensive and impractical. This paper proposes an automatic passenger detection device based on video image analysis. This device extracts and identifies target individuals from collected video images, enabling passenger counting. This invention overcomes the limitations of infrared sensors in passenger counting, improves data accuracy, and further optimizes proactive, energy-efficient, and highly effective control strategies. The research has significant practical and social implications.
The device is mainly composed of three modules: video image acquisition, image processing and data result output [6]. The video image processing module is the core module of the detection and statistics device, and its performance determines the main working performance indicators such as the overall processing speed and accuracy of the device. The background, decoration, lighting and color of the waiting hall are complex and diverse due to different architectural styles. The objects are dynamic and static. Therefore, it is necessary to use appropriate processing algorithms to remove the background, update the background in time, and extract the target by segmenting the foreground edge. The waiting area is divided into zones, and a hemispherical wide-angle pan-tilt integrated camera with a large coverage area and a CCD lens is used. When the density of people waiting for the elevator is greater than the set threshold, the camera lens is automatically adjusted to maintain a certain degree of redundancy and overlap with the sensitive area to improve the video image acquisition rate and reduce the error. A smart decoder with anti-crash performance and supporting DC12V/AC24V output automatic matching control protocol is selected; the output end is transmitted to a video server that realizes various special interactions and frame-by-frame retrieval functions for encoding, compression and transmission; the host computer receives the acquired video data through the antenna and supports it through software. Image processing based on inter-frame information of dynamic sequence video images is all grounded in motion detection, which is also the first step of the core image processing module of the system. Motion detection processing based on background subtraction is chosen to completely separate the foreground and background information of the image. Threshold background-foreground segmentation is selected to solve the shadow problem and the perturbation error in the processing, so as to completely remove the background information and extract the complete moving target information in the current frame. Under the processing of a combination of complex algorithms, the background and human limb models are extracted, and the number of people is determined. The numbers are automatically counted and stored, and the data on the number of people waiting for the elevator is actively sent out. HTTP/TCP protocol is used for transmitting control information, and RTP/UDP protocol is used for transmitting real-time data. The workflow of the detection device is shown in Figure 4.
Figure 4. Flowchart of the elevator occupancy detection device
5. Conclusion
In the Matlab environment, simulations of a twin elevator group control system were conducted under three typical traffic modes: upward, downward, and inter-floor. These simulations employed traditional particle swarm optimization (PSO), adaptive mutated PSO, hybrid PSO, and the traditional minimum waiting time algorithm, all with varying passenger flow data. The results, based on the average passenger waiting time, long waiting time incidence, and number of stops obtained from the simulations, demonstrate that the two improved PSO algorithms achieve better control of the target, validating the effectiveness and applicability of the algorithms. Furthermore, the twin elevator group control system can utilize different control algorithms according to different traffic modes, thereby improving the elevator's service quality and operational efficiency. Since research on twin elevators is still in its early stages worldwide, this study primarily focuses on the theoretical level, and its application in engineering practice requires further verification.
References
[1] Zhang Jin. Research on Twin Elevator Group Control System Based on Particle Swarm Algorithm [D]. Harbin Institute of Technology, 2014.
[2] Wang Suhua. Design and Implementation of a New Type of Elevator Traction Machine Drive and Control System [D]. Nanjing University of Science and Technology, 2013.
[3] Li Qingchao. Research on composite control system of twin elevator groups based on genetic algorithm [D]. Harbin Institute of Technology, 2013.
[4] Huang Zhaobin. Research on performance optimization method of permanent magnet synchronous elevator traction system [D]. South China University of Technology, 2014.
[5] Liu Rong. Research on Adaptive Particle Swarm Optimization Algorithm and Its Application in Multi-Objective Optimization [D]. South China University of Technology, 2011.
[6] Lin Qin. Analysis of large-scale crowd density and abnormal behavior in videos [D]. Xiamen University, 2014.
About the author: Zhang Jinyang (1976-), female, master's degree, senior engineer.
