Neural network does not depend on the model of the control object. It can approach the dynamic behavior of the unknown non-linear object by learning, so it can be used in the control of three-level PWM rectifier. However, due to the requirement of fast and real-time performance of three-level PWM rectifier, the practical application of neural network in this field is still limited. This paper studies the current control method based on single neuron, gives the current control structure based on single neuron, designs a single neuron current controller, and finally carries out simulation experiments.
The results show that the three-level converter with single neuron current controller has fast response speed, almost no overshoot, and can quickly restore to stable state when the load changes abruptly. Single neuron is the basic structure of neural network. It has the characteristics of simple structure and small computation.
As a controller, the dynamic performance of the system depends only on the error signal, and is not affected by the parameters of the object model, so it can improve the performance and robustness of the system. Single neuron controller combines the advantages of PID control, can adjust the parameters of PID on-line, has the ability of self-adaptation and self-learning, and can meet the requirements of speed and real-time, so it has been widely used in various control systems. The single neuron adaptive controller composed of single neuron is used for current control of three-level rectifier. The structure of current control link of the system is shown in Figure 1.
In the control system of three-level rectifier, single neuron current controller is used to control current and current. Its structure is shown in Fig. 2. Current control consists of three parts: single neuron controller, current detection and coordinate transformation.
Fig.
2 shows the structure of single neuron current controller for axis current. The input of the controller is the difference between the given value of axis current and the actual feedback, and the output is the component of axis control voltage. In the optimal control theory, the quadratic performance index is used to calculate the control law to obtain the desired optimization effect.
In the learning algorithm of neuron, the idea of quadratic performance index in optimal control can also be used to introduce quadratic performance index into the adjustment of weighting coefficient. By minimizing the sum of weighted square of output error and control increment, the weighting coefficient can be adjusted, thus indirectly realizing the constraint control of output error and control increment weighting. 。 Among them.
It is the Euclidean norm of the weight vector. Dividing by norm means that the weight vector is united in the weight vector space to ensure the convergence of the control strategy. In the above formula, P and Q are the weighted coefficients of output error and control increment, and the reference input and output at k time respectively. In the formulas above, thermostatic element the learning rates are corresponding to the weights, and K is the total learning rate.
The simulation model of single neuron controller based on SIMULINK is built, and the simulation model of PWM and rectifier is studied. The simulation model of single neuron controller is shown in Figure 3. Grid parameters: Em = 311V, f = 50Hz; AC side parameters: Ls = 6mH, Rs = 0.5; DC side parameters: Cd = 2200uF, L0 = 3mH; DC side voltage Vdcg = 600V; switching frequency FS = 2KHz. When the rectifier is started, it is equivalent load, output power is 18KW, reactive current, and the rectifier works under the condition of unit power factor. When the system is stable for 0.
3 seconds, the equivalent load of 18KW is connected in parallel on the DC side. The DC side voltage and AC side current and voltage waveforms are shown in Figure 4. The simulation results show that when the load changes abruptly, the DC side voltage of the three-level converter with single neuron current controller restores to the steady state value after 0.13 seconds, and the voltage value restores quickly. The current waveform of the power grid is sinusoidal and keeps in phase with the voltage of the power supply. The rectifier still works under the condition of unit power factor. In this paper, the performance of rectifier under sudden load change is studied through simulation experiments. The three-level rectifier with single neuron current controller has fast response speed, almost no overshoot, and can recover to stable state quickly when the load changes abruptly.