Fan fault detecting method based on depth belief network model

Abstract

The invention discloses a fan fault detecting method based on a depth belief network model. The fan fault detecting method comprises that a fan flow forecast model is established by using depth beliefnetwork, state parameters related to the fan flow are selected to be used as the input variable of the model, and the flow of a fan is forecasted. The depth belief network is formed by staking multilayer continuous type limit Boltzmann machines, and an adaptive step length method is used to accelerate the algorithm training process. In addition, residual distribution characteristics are calculated according to a sliding window, when the mean value or standard deviation of the residual exceeds the threshold, alarm is issued. By means of the fan fault detecting method based on the depth beliefnetwork model, the flow of the fan can be accurately forecasted, meanwhile, the abnormal working state of the fan can be detected, and fault detection on the fan is achieved.

Description

technical field [0001] The invention belongs to the field of thermal automatic control, and relates to a fan fault detection method, in particular to a fan fault detection method based on a deep belief network model. Background technique [0002] The fan is very important to ensure the safe and reliable operation of the entire power generation system, and it is the equipment that needs to be monitored during the operation of the unit. [0003] With the demand for peak-shaving capacity of the power grid and the increasing installed capacity of generators, wind turbines are often in the operating state of high parameters, large vibration, poor working conditions, and fast load adjustment. The reliability of wind turbine operation decreases and the failure rate increases. In addition, the structure of the fan is complex and the system is highly nonlinear, so it is difficult to establish an accurate analytical mathematical model of the equipment. [0004] With the development o...

AI Technical Summary

Problems solved by technology

Neural network and support vector machine are two relatively mature algorithms, but the traditional neural network training method is complex and easy to fall into local optimum

Support vector machine is difficult to implement for large-scale sample training, it requires solving the quadratic programming problem under inequality constraints, and the operation time is long

Images