Application Of An Improved Particle Swarm Optimization Algorithm In Elevator Fault Diagnosis

Since the new century,the economy has developed rapidly,the people's living standard is also increasing,and more and more high-rise buildings appear in people's daily life.Elevator has also become an indispensable and important transportation tool in daily life.According to statistics,in 2017,the total output of elevator in China totaled 679 thousand units,an increase of 5.1% over the same period.According to the prediction of the special equipment Bureau of AQSIQ,it is estimated that the total elevator output of China will be 725 thousand and 300 units in 2018,an increase of about 6.8 percentage points over the same period.In the face of such a huge demand for elevators,people are enjoying the convenience brought by the elevator,and they are also worried that the frequent occurrence of elevator failures will threaten people's property and even life.In order to reduce elevator failure,besides ensuring product quality and regular maintenance,it is more important to have an effective elevator fault diagnosis system.In recent years,the neural network technology has developed rapidly.It has a strong nonlinear fitting ability,and can map any complex nonlinear relation,and the learning rules are simple and easy to be realized by computer.Therefore,this paper mainly studies the method of elevator fault diagnosis based on neural network.The design ideas of this paper are as follows: first,the data that can reflect the characteristics of the elevator fault is collected,and the input data of the neural network can be used as the input data of the neural network.One part of the data is selected as the training data of the network and the other part of the network test data.Secondly,probabilistic neural network is used as the model of elevator fault diagnosis,and the improved particle swarm optimization algorithm is used to optimize it.Finally,the output of the model is compared with the correct structure to determine the accuracy of the model for elevator fault classification.In order to establish a neural network based elevator fault diagnosis system,the following works are done in this paper.(1)The selection of neural network: the most commonly used neural networks are BP neural network,RBF neural network,grey neural network and so on.In many neural networks,probabilistic neural network has the advantages of simple process,fast convergence speed and strong sampling ability.Therefore,in this paper,probabilistic neural network is used as the model of elevator fault diagnosis.(2)Elevator fault data acquisition: in the elevator operation,it mainly involves 4 systems,such as mechanical system,control speed control system,security system,power system and so on.Therefore,this paper collects the elevator fault data from the above 4 systems.It mainly collects wire rope slip distance,safety clamp action lifting force,speed limiter speed,brake pull force,up and down current,starting and braking maximum speed.In order to make all the data in the [0,1] range,this paper adopts the data normalization processing method,and obtains 94 sets of fault data.80 groups are used as the training data of the probabilistic neural network,and the remaining 14 groups are used as the test data.(3)Optimization of probabilistic neural networks: in previous studies,the smoothing factor of probabilistic neural networks usually uses empirical values.In order to overcome one of the problems,this paper uses particle swarm optimization to optimize the smoothing factors of probabilistic neural networks.However,particle swarm optimization algorithm is prone to fall into local extremum,so we need to improve particle swarm optimization.Among many improvement strategies,a classical improvement strategy is an improved strategy of decreasing inertia weight.However,there are serious defects in this method: the particle search ability in the later stage of the iteration becomes weaker,which makes the particles fall into local optimum.In this paper,a strategy of sinusoidal change of inertia weight is adopted.Because the sine function is a typical periodic function,the improvement strategy can be used to synchronize the global optimization of particles with their own optimization.Finally,the improved strategy is compared with the traditional particle swarm optimization algorithm and the inertia weight linear decreasing strategy.The superiority of this method is verified.(4)Elevator fault diagnosis: using improved particle swarm optimization algorithm to optimize the smoothing factor of probabilistic neural network,as a model of elevator fault diagnosis.At the same time,a comparison is made between the smoothing factor and the linear decreasing inertia weight.It is verified that the accuracy of this method is higher in elevator fault classification.