Dynamic Load Identification For Turbine Generator Set Based On Neural Network

Vibration in turbine generator set and powerhouse is ubiquitous because of their structure speciality and function. The dynamic load on powerhouse when operating is the basic data of dynamic design and vibration analysis. However, because the generator set is so large-sized and the load distribution is very complicated, it is difficult to measure dynamic load directly. So, It is significant that using load identification technique to identify dynamic load. In this paper, the author tries to identify dynamic load of generator set by neural network.The neural network is a new method about load identification, it is superior to traditional methods in many aspects. By this method, there is no need to calculate modal matrix, stiffness matrix and mass matrix, the method is also easy to learn, and has high precision.In the first part of this paper, the author compares identification results calculated by different neural network algorithms. At first, designing network structure of BP network based on momentum ways and BP network based on LM algorithm and RBF network. Then applying three networks into a numeric example to identify different dynamic load and comparing the load calculated by three algorithms. The result illustrates that BP network based on LM algorithm is superior to the other two algorithms in identifying load.In order to validate the identification effect based on LM algorithm, in the second part, the author discribes indoor model experiment about dynamic load identification and analyses the result. In the experiment, the author makes a simple frame structure simulating upper part of powerhouse, then places accelerate sensors on model according to result calculated by optimization layout of sensors. By this experiment, getting acceleration response and translating it into displacement response, then putting it into network which has been trained. The result illustrates that identifying dynamic load by BP network based on LM algorithm is effective. In the last part of this paper, the author builds fern model of generator set according to the dynamic characteristic and identifies dynamic load on turbine and stator. Result illustrate that identify dynamic load of generator set is feasible.