Original link:tecdat.cn/?p=6585
In this paper, the design of hybrid fuzzy sliding mode loss minimization control for permanent magnet synchronous generator (PMSG) and high-performance online training recursive neural network (RNN) is introduced. The back-propagation learning algorithm is used to adjust the RNN controller. PMSG speed uses maximum power point tracking below rated speed, which corresponds to low and high wind speeds, and can capture maximum energy from the wind. A sliding mode controller with integral operation switching surface is designed and the upper bound of uncertainty is estimated by fuzzy reasoning mechanism.
Introduction to the
Recently, wind power systems have attracted great attention as a clean and safe renewable energy source. Wind power can be operated by constant speed and variable speed operation using power electronic converters. Variable-speed systems are more attractive than fixed-speed systems because of improved wind power generation and reduced flicker problems. And wind turbines are able to operate at maximum power points with respect to various wind speeds, optimally achieving maximum efficiency at all wind speeds by adjusting shaft speed.
Design of RNN variable propeller control system
Architecture of RNN
RNN has three layers, namely, input, hide and output. The proposed RNN is used for turbine controller in the three-layer neural network diagram shown.
Figure. Architecture of RNN.
matlab
clc; clear; close all; setup; dbstop if error %% RNN % useGPU = true; dataType = 'single'; backward = true; input = InputLayer(struct('batchSize',5,'useGPU',useGPU,'dataType',dataType,'backward',backward,'reverse',true)); embedd1 = EmbeddingLayer(struct('hidden_dim',512,'input_dim',input.vocabSize,'useGPU',useGPU,'dataType',dataType,'backward',backwa rd)); rec1 = RecurrentLayer(struct('hidden_dim',512,'input_dim',512,'useGPU',useGPU,'dataType',dataType,'backward',backward)); rec2 = RecurrentLayer(struct('hidden_dim',512,'input_dim',512,'useGPU',useGPU,'dataType',dataType,'backward',backward)); loss = SoftmaxLayer(struct('hidden_dim',input.vocabSize,'input_dim',512,'useGPU',useGPU,'dataType',dataType,'backward',backward )); % train MaxIter = 10000; history_cost = zeros(1,MaxIter); for i = 1 : MaxIter tic; target = input.fprop(struct('reverse',false,'fprop',true)); loss.fprop(rec2.fprop(rec1.fprop(embedd1.fprop(target,size(target,2)),size(target,2)),size(target,2)),size(target,2)); history_cost(1,i) = gather(loss.getCost(target)); display(history_cost(1,i)); embedd1.bprop(rec1.bprop(rec2.bprop(loss.bprop(target)))); loss.update(@SGD); rec2.update(@SGD); rec1.update(@SGD); embedd1.update(@SGD); toc; endCopy the code
The simulation results
The traditional proportional integral (PI) controller is widely used in industry because of its simple control structure, easy design and low cost. However, if the controlled device is highly nonlinear or the desired trajectory changes with higher frequencies, the PI controller may not have satisfactory performance. For variable-speed wind turbines, mechanical actuators are usually used to change the pitch Angle of blades to control the power coefficient.
Dynamic performance test
Figure (a) shows the shaft speed; The system can be observed to track the maximum power at rated generator speed. Figure 6 (b) shows the verification of maximum power tracking control.
FIG. Simulation results of wind speed profile :(a) wind profile speed tracking, (b) maximum power tracking control signal, (c) power tracking error and (d) power coefficient c p.
Efficiency performance test
The graph shows the efficiency performance. Consider that the converter loss in the medium-high speed range is almost 80%.
Simulated efficiency performance:
conclusion
This paper introduces the design of driving PMSG system for variable speed wind energy system. The speed controller sets the generator torque instruction, which is achieved through the current control loop. The proposed system has been implemented in real-time applications with commercial PMSG and DC drivers that simulate the behavior of wind turbines. Simulation results show that the system has good performance.
Dynamic performance shows accurate estimation of speed not only at steady state but also at fast input, while traditional PI controller cannot determine the uncertainty model of various wind speeds. In this study, a fuzzy sliding mode velocity and RNN pitch controller were successfully implemented to control WECS. This technique can maintain the system stability and achieve the desired performance even if the parameters are uncertain.