INTRODUCTION
Electric power generation using renewable energy presents tremendous social, economic and environmental benefits over fossil and nuclear power generation. Further more, renewable energy enables the development and applications of a new mode of power generationdistributed power generators as a complement to traditional central power generation stations. The area of west of China is far more that of east of China, but the terrain of west is very complicate. Electrical energy wasn’t transported any isolated areas of west. However, numerous isolated areas have significant wind and solar energy potential. If these resources should be used effectively in isolated areas, the problem of absence of electrical energy will be resolved. Though there are so many advantages for renewable energy, there are uncertainty and nonline character for power of renewable energy. The unstable factors of these energy results in the power quality of a distributed power generation system that is bad. Wind and solar energy and storage energy equipments are combined to stabilize the distributed power generation system^{[16]}. The storage energy equipments perform the storage and release process of energy and smooth the DC bus voltage in the system. The stability problem of a system with DC bus voltage 650 V is introduced^{[7]}. In this study DC bus voltage of system is less than 200 V. The system is based on a vectorcontrolled induction machine driving a flywheel and addresses the problem of regulating the DC bus voltage with neural network PID controller.
STRUCTURE AND PROMCIPLES OF THE DISTRIBUTED POWER GENERATION SYSTEM
The present study primarily focused the studies of the energy control of the wind energy system (Fig. 1). The wind power generator is implemented by a simulator unit composed of an induction motor and a permanent magnet synchronous generator. The induction motor simulates the characteristics of a wind turbine under variable wind speeds and drives the permanent magnet synchronous generator to generate electric power. As one of storage units, the flywheel storage equipment is composed of an induction motor and steel flywheel.

Fig. 1: 
Structure of the distributed power generation system based
on renewable 
Due to the power of wind generation system is changed with the wind speed and load is changed with the time, DC bus voltage fluctuates, that is the system is unstable. When the DC bus voltage E_{dc} decreases, the induction machine is controlled to operate as a generator, transforming the inertial energy stored in the flywheel into electrical energy supplied to the DC Bus. When the DC bus voltage E_{dc} increases, the induction machine motors, transferring energy from the DC Bus to the flywheel.
SYSTEM MODELLING AND NEURAL NETWORK CONTROLLER CONSTRUCTING
In the system, it is based on a standard indirectrotorfluxorientated (IRFO)^{[8]} control of the induction machine driving the flywheel. The dq current and voltage values are referred to the reference frame aligned to the rotor flux and take DC values in steady state (Fig. 2).
The torque current reference I^{*}_{q} is derived from the E_{dc} fuzzy controller.
Mathematical model of energy control system: The mathematical model
of the system is built according to the principles of the energy control. In
Fig. 2, the power balance between the DC bus side and the
induction machine side is expressed as:
Where, E_{dc} is the DC bus voltage, i_{G} is the output current of wind generation simulator, i_{L} is the current of DC load, P_{loss} is the inverter and iron power losses, k is the coefficient of the 23 axes scaling, V_{d} and V_{q} are the dq voltage of the stator respectively and i_{d} are the i_{q} current of the stator, respectively, C is the total capacitance of the DC bus.
According to the mathematical model of rotorflux orientation of induction machine^{[2]} , V_{d}, i_{d}, V_{q} and i_{q} are given by:
Where, R_{s} and R_{r} are the stator and rotor resistance,
respectively, L_{s} and L_{r} are the stator and rotor inductance,
respectively L_{m} is the magnetizing inductance, φ_{r}
is the rotor flux, ω_{s} is the rotational speed of stator, p is
the differential operator,

Fig. 2: 
Energy control system based on induction machine 

Fig. 3: 
PID controller structure based on BP neural nerwork 
Using (1)(3), it can be shown that the power balance can be de rived as:
Since the flywheel inertia will be large (the speed dynamics will be slow) and neglecting the variation in the energy stored in σL_{s}, C and P_{loss} then the steadystate system equations may be used effectively. Using Eq. 4, the steadystate power balance is obtained as:
Using Eq. 5, the steadystate torque current i_{q} is expressed as:
Neural network PID controller constructing: According to Eq. 6, the structure of neural network PID controller for the DC bus voltage E_{dc} is shown in Fig. 3. The input current i_{G}i_{L} of the flywheel storage energy system is obtained by the fuzzy control rule of DC bus voltage E_{dc}.
According to state change of the system, the parameters of PID controller (K_{P}, K_{I} and K_{D}) are regulated by the BP neural network (Fig. 3). There are three layers in the BP neural network (Fig. 4). It is composed of input layer, hide layer and output layer. The activated functions (f(x) and g(x)) of hide layer and output layer are expressed as:
The operation process of neural network PID controller is composed of working and learning. Every weight of nerve cell is not change in the working process. According to input signals, the outputs (K_{P}, K_{I} and K_{D}) are computed by the neural network and then they are delivered to the PID controller in order to obtain the value of i_{G}i_{L}. Every state of nerve cell is not change in the learning process. In order to obtain the desired output character of the neural network, the weight values between nerve cells are regulated according to learning sample. So the neural network PID controller can regulate parameters online with the change of object. It takes on the adaptive ability.
EXPERIMENTAL RESULTS
In the present study, induction machine of rating power 550 W is used for the
actuator of the flywheel storage energy system. PMSM of rating power 2 kW and
induction motor of rating power 2.2 kW are combined for the actuator of the
wind generator simulator. The parameters of 550 W induction machine are shown
in Table 1. The parameters of 2 KW PMSM are shown in Table
2.
Table 1: 
Parameters of the induction motor 

Table 2: 
Parameters of the PMSM 


Fig. 5: 
Schematic diagram of experimental system 

Fig. 6: 
The response of current i_{G}i_{L} and DC
bus voltage E_{dc}, connected at t=0.36 sec and disconnected at
t=0.79 sec 
Figure 5 shows the Experimental system, which is composed
of the control structure of Fig. 3 embedded in the system
structure of Fig. 2 and wind generation simulator.

Fig. 7: 
The response of DC bus current I_{dc} and DC bus power
P, connected at t=0.36 sec and disconnected at t=0.79 sec 

Fig. 8: 
Experimental system. (In the picture: 1wind generation simulator
unit including an induction motor and a permanent magnet synchronous generator;
2DC load; 3flywheel energy storage unit; 4converters and control system;
5control computer) 
The picture of experimental system is shown in Fig. 8. A
incremental encoder provides the flywheel speed. The demanded DC bus voltage
is 120 V. The demanded load is 1500 W.
Figure 6 shows the response of current i_{G}i_{L} and DC bus voltage E_{dc}. Figure 7 shows the response of DC bus current I_{dc} and DC bus power P. The load is applied at t=0.36 sec and disconnected at t=0.79 sec. When the load is connected, the DC bus voltage E_{dc} moves towards down of the demanded voltage. The output current of the neural network PID controller also moves towards down, that is the load current is more than the generation current. At the same time, the DC bus Current values are negative; the stored energy is delivered from the flywheel to the DC load. Finally, the DC bus voltage E_{dc} moves towards the demanded voltage. When the load is disconnected, the DC bus voltage E_{dc} moves towards up of the demanded voltage. The output current of the neural network PID controller also moves towards up, that is the load current is less than the generation current. At the same time, the DC bus Current values are positive; the flywheel stores the redundant energy generated from wind. Finally, the DC bus voltage E_{dc} moves towards the demanded voltage.
CONCLUSION
This study has proposed a new energy control strategy of a distributed power generation system. The control strategy for regulating the DC bus voltage is performed in an experimental wind flywheel hybrid energy system. The system uses a metal flywheel for power smoothing employing an inverterfed vectorcontrolled induction machine. The neural network PID controller embedded in the system is used for the object of energy control. The experimental results confirmed the satisfactory operation of the proposed system.