Search

The importance of short-term wind power forecasting is significantly increased because of the demand of green energy and large-scale integration of the wind power plants in the electric network. In this paper, a Gaussian mixture model (GMM)-based radial basis function neural network is proposed to forecast the short-term wind power generation. Actual measured wind power output data are adopted to implement the proposed model. Test results of wind power obtained by autoregressive integrated moving average (ARIMA), back propagation neural network (BPNN), radial basis function neural network (RBFNN), support vector regression (SVR), and the proposed method are then under comparisons. Simulated results show that the presented method leads to more accurate wind power forecasting.

In conventional power systems, client-server architecture and centralized control are adopted for data acquisition and supervisory control, and the speed of message transmission and the fault responses is slow. In contrast, various types of distributed energy resources (DERs) with different characteristics and of different capacities exist in a microgrid system, and client-server architecture and centralized control are inadequate for control and operation purpose. In this study, an example microgrid system with a multi-agent system which complies with FIPA (Foundation for Intelligent Physical Agents) standard is proposed and built using MATLAB/Simulink and Java Agent Development Framework (JADE). The operations of and among the intelligent electronic devices (IEDs), circuit breakers (CBs), calculation agent and CB agents in the example microgrid system for fault protection coordination under both normal and short-circuit fault conditions are analyzed and discussed. The results demonstrate that protection coordination can be correctly planned and control actions can be executed properly by the multi-agent system in the microgrid system, and the multi-agent system can isolate the fault area and protect the microgrid system in real time under varies conditions.

With the increasing use of non-linear loads, the harmonic problems have become of great concerns. One of the solutions to harmonics is the application of shunt active power filters (APFs). The APF reference compensation strategy proposed in this paper adopts the backpropagation neural network (BPNN) to train the control algorithm and calculate the reference compensation current under steady state and varying load levels. Simulation results obtained by using MATLAB/Simulink show that the proposed BPNN-based APF control strategy can effectively mitigate harmonic currents generated by the nonlinear load.

The demand of renewable energy resources has become an important issue for green power generation in many countries. With the increasing photovoltaic generation in the distribution system, voltage fluctuation problems often present in the PV-connected grid affected by the moving cloud shadows. This paper presents a study to evaluate impacts on voltage quality due to cloud shadows movement over high-penetration PVs in a distribution power system in southern Taiwan. The moving-cloud shadow model is built under Matlab/Simulink incorporated with OpenDSS and is applied to show voltage fluctuations. Results show that the developed simulation tool is suitable for assessment of the PV output versus voltage changes due to cloud moving shadow impacts.

A division-summation (D-Σ) digital control for three-phase inverters to achieve active and reactive power injection has been presented. It can overcome the limitation of abc to dq frame transformation. However, the D-Σ control requires four sets of control laws to cover four quadrant operations: grid connection (PF 1 ∼ ±0.866), rectification with power factor correction, PF leading (0 ∼ 0.866), and PF lagging (0 ∼ −0.866). Moreover, the switching sequences for the four modes are also different from each other, increasing complexity of firmware programming. In this paper, improvement of the control-law derivation and region selection for the D-Σ digital control is presented. By selecting the zero crossing points of phase voltages as region transitions, the control laws and the related parameter tables for the four modes can be unified to a general form. The switching sequences of the four modes are also unified to the switching pattern of the grid-connection mode. Additionally, a D-Σ transformation matrix is identified to simplify the derivation procedure of the division (D) and summation (Σ), which can obtain the control law directly. Measured results from a 10 kVA 3Φ bi-directional inverter have been presented to confirm the improvement.