Study On Passenger Flow Analysis And Optimization Control Methods For Modern Elevator Traffic System

Elevators, served as the most essential transportation junctions in high-rise buildings, attract more and more attentions on their safety, effectiveness, comfort and reliability. Because of the environmental complexity and the operation dependency between different elevators, traditional elevator control system is not capable to adjust the variation of passenger flow or to judge the unsafe factors around. Based on new theories and methods in the subjects of computer recognization and artificial intelligence, some critical questions are discussed and analyzed in this thesis, including passenger flow data retrieval, status analysis, traffic model recognization and optimal control strategy. A new solution is therefore proposed to achieve high effectiveness of modern elevator operation. The main work done is as follows.First, the key factors of passenger flow description in elevator transportation system are analyzed, and the issues of object dectection and feature capturing are studied based on depth information. Then an algorithm is proposed to recognize elevator passengers based on Hough Forests. The Hough Forest decision trees are built up, and the voting probability of object detection is increased through Hough transform. The experienmental results show that this new method is capable of detecting and locating human bodies, and reduces the error rate induced by insufficient lighting.Second, in order to reflect the variation of elevator passenger flow, the technique of passenger tracking is needed. The local block-graphs targets matching (LBTM) is studied, which conducts local image processing and associated data analysis to predict human body gestures. Based on HOQ an algorithm is raised to combine image processing and depth information for human body tracking. According to the experienments in various circumstances and result anaylsis, tracking effectiveness is highly increased with the help of the depth information techniques.Third, the methodology of human behavior prediction and reconstruction is studied. Based on the kernel method, human behavior samples in low-dimensional space are mapped into high-dimensional space. Through the algorithm of kernel ridge regression, unknown data are regressed and human behaviors can be predicted. Then, an algorithm based on Locally Linear Embedding (LLE) and Kernel Ridge Regression (KRR) is proposed. The method uses the image color and depth information to extract the human skeleton and established training set of action, and turns the training set into action vectors library, and evaluates the human action of low-dimensional manifold according to LLE algorithm. Finally, the prediction action points of low-dimensional are mapped back to higher-dimensional Euclidean space to reconstruct actions. The experimental result is effective enough to show that the proposed method is capable of predicting elevator passenger behaviors and therefore has the capability to judge the unsafe factors inside elevators and to do early-warnings.Next, the features of various elevator traffic models are analyzed, and a way to recognize traffic models in elevator group controlling system is proposed based on Random Forest. In this method, decision trees are firstly constructed by using Random Forest algorithms, and then through weighted voting, the current tranffic model can be recognized. The simulation results show that the proposed method is effective in elevator traffic model recognition.Last, in order to optimize the dispatched control of elevator transportation system, a group of performance indicators is considered including passenger waiting time, journey time, system energy consumption, fitness and so on. And then, a fitness-evaluational function is created for multi-objective optimization. In addition, the improvement of genetic optimization algorithms is also studied, and an algorithm, PSO-GA, is proposed, which combines Particle Swarm Optimization (PSO) and Genetic Algorithms (GA) together. Furthermore, another algorithm, GNP, is also proposed, which is based on Genetic Algorithms (GA) and Genetic Planning (GP). GNP expands the expression capability by using directed structure graphs. The simulation results show that the improved algorithms have great optimization capability and are feasible to optimize the dispatched control of elevator transportation system.