Abstract: This paper proposes a novel application method and sample acquisition approach for artificial neural networks in switched reluctance motor (SRM) control systems. A Simulink-based simulation system is built and applied to SRMs with severe nonlinearity. The characteristics of the control system are improved by combining the advantages of parameter-adjustable neural network control and variable learning rate algorithms. Experimental results show that the proposed control system has advantages such as fast dynamic response, small overshoot, good robustness, and significant energy saving.
Keywords: Neural network; Switched reluctance motor; Simulink; Nonlinearity
Intermediate Classification Number : TP9 Document Identification Code: B
0 Introduction
A switched reluctance motor (SRM) is a multivariable, strongly coupled nonlinear system. Its parameters and structure vary under different control methods, making it difficult to achieve satisfactory results using traditional PID control. Neural network technology can be used to solve the control problems of complex systems that are difficult to control effectively using traditional methods. Intelligent control can fully utilize its nonlinearity, variable structure, and self-optimization capabilities to overcome the variable parameters, variable structure, and nonlinearity of the SRM speed control system, thereby improving the robustness of the entire control system. Therefore, introducing intelligent control into a SRM will help improve the system's performance indicators.
The research and application of neural network control have penetrated into numerous fields and disciplines. Similarly, its development has brought new ideas and methods to the control strategies of electrical drive systems. Compared with traditional control methods, neural network control has a series of advantages: Firstly, it breaks through the traditional control theory's requirement to be based on the mathematical model of the controlled object; it controls according to the actual effect without relying on the mathematical model of the controlled object. Secondly, intelligent control mimics human brain thinking, and since human brain thinking is nonlinear, neural network controllers also possess nonlinear characteristics.
Currently, AC and DC drive systems have relatively mature control schemes, such as DC dual closed-loop control and vector control systems for AC motors. Furthermore, after inner-loop modifications (current loop, vector transformation), the speed loop of AC and DC electric drive systems can be modeled using a unified mathematical model, and traditional PID control can achieve basically satisfactory control results. However, in actual drive systems, the parameters of the motor itself and the parameters of the driven load may change during operation. Therefore, electric drive systems are essentially time-varying, strongly coupled, nonlinear, multivariable systems. The parameter variations and nonlinear characteristics of the controlled object mean that fixed-parameter PID controllers cannot maintain the design performance indicators under various operating conditions, resulting in weak system robustness. Introducing neural network control into electric drive systems can utilize its nonlinearity, variable structure, and self-optimization capabilities to overcome the variable parameters and nonlinear factors in electric drive systems, thereby improving system performance.
1. Basic Structure and Operating Principle of Switched Reluctance Motor
The stator and rotor of a switched reluctance motor have salient pole structures. Each tooth of the stator core has a concentrated winding, similar to the main pole winding of a DC motor, while the rotor core teeth do not have windings. The windings on two opposite teeth on the inner circumference of the stator are connected in series (in the same direction) to form one phase winding. Figure 1 shows a schematic diagram of a four-phase 8/6-pole switched reluctance motor.
Figure 1. Schematic diagram of a four-phase 8/6-pole switched reluctance motor.
Because switched reluctance motors (SRMs) have a doubly salient pole structure, there are many combinations of the number of poles in the stator and rotor. The most common configuration is the four-phase 8-pole stator and 6-pole rotor configuration shown in Figure 1. Other configurations include a three-phase 6-pole stator and a 4-pole rotor, or a three-phase 12/8-pole configuration. Considering that increasing the average rate of change of stator phase winding inductance with rotational angle can improve motor output, the most common configuration is that the rotor has two fewer poles than the stator. Let the number of phases of the switched reluctance motor be denoted as m, the number of stator poles as Ns, the number of rotor poles as Nr, and the step angle as θs. Then:
Generally, the higher the number of phases in a motor, the more teeth the stator and rotor have, the smaller the step angle, and the smaller the output torque ripple. However, increasing the number of phases also brings a series of problems, such as increased complexity of the power converter's main circuit, an increased number of power devices, and increased cost. Therefore, single-phase and two-phase structures are more commonly used in applications with lower requirements, while switched reluctance motors used as drives often adopt three-phase or four-phase radial structures.
Switched reluctance motors operate based on the reluctance effect, following the "principle of minimum reluctance." This principle states that magnetic flux always closes along the path of least reluctance. In a magnetic field, the main axis of a given iron core tends to move towards the position coinciding with the magnetic field axis. Utilizing this tendency, the switched reluctance motor generates a magnetic field with stator salient poles, and the rotor core salient poles form multiple evenly distributed main axes. By controlling the sequential generation of magnetic fields in each phase of the stator, the rotor always tends to move towards the position of minimum reluctance, thus generating continuous torque to maintain motor operation. Unlike synchronous motors, where the rotor position is entirely determined by the current conduction, this magnetic field generated by controlling the phase current is not entirely determined by the current conduction. Instead, the current conduction depends on the rotor position. Obtaining rotor position information is mostly achieved using position sensors.
2 Neural Network PID Controller
The structure of the PID control system based on neural network tuning is shown in Figure 2. The controller parameters are optimized by adjusting the network weights. A variable learning rate is used to accelerate the network convergence speed. The RBF online identification network identifies the online parameters of the switched reluctance motor and adjusts the controller parameters in real time according to the torque changes.
Figure 2. Structure diagram of PID control system based on neural network tuning
2.1 Structure of Fuzzy Neural Networks
The fuzzy neural network has four layers, as shown in Figure 3. Layer 1 is the input layer; layer 2 is the fuzzification layer; layer 3 is the fuzzy inference layer; and layer 4 is the output layer. The fuzzy neural network structure is 2–6–6–3.
Figure 3. Structure of a fuzzy neural network
(l) Input Layer. This layer takes the input error e and the actual system output y(k) as inputs to the next layer. The activation function is:
Therefore, the outputs of this layer are e and y(k).
(2) Blurring layer. The activation function is the membership function. Therefore, the output is:
Where i = 1, 2; j = 1, 2, ..., 6. cij and bij are the mean and standard deviation of the membership functions of the j-th fuzzy set of the i-th input variable of the Gaussian function, respectively.
(3) Fuzzy Inference Layer. The output value of this layer is obtained by multiplying the fuzzy quantities in the upper layer pairwise. Therefore, the activation function of this layer, i.e., the output, is:
Here, k = 1, 2, 3, 4, 5, 6.
(4) Output Layer. This layer outputs the parameters of the PID controller. The output value of this layer is the weights multiplied by the output of the third layer using a matrix multiplication method. Therefore, the output of this layer is:
3. Experimental Simulation
The dynamic response characteristics of a switched reluctance motor controlled by a neural network were studied experimentally. The experimental system used a four-phase 8/6-pole switched reluctance motor with a rated power of 60 kW and a rated speed of 1500 r/min. Figure 4 shows the system response process under the control strategy proposed in this paper. Figure 5 shows the online tuning of the PID parameters using the fuzzy neural network.
Figure 4. System dynamic response curve (using the method described in this paper)
Figure 5. Online tuning of PID parameters using a fuzzy neural network.
4. Conclusion
Neural network controllers are particularly suitable for adaptive control of nonlinear objects. This paper combines fuzzy theory, system identification, and neural networks, and utilizes their control method of adjusting PID parameters to achieve excellent control performance in a PID control system, solving the problems encountered by ordinary PID controllers in controlling time-varying and nonlinear systems. Experiments using a MATLAB simulator demonstrate that the system exhibits good dynamic characteristics, as well as good adaptability and robustness.
