Industrial engineering, machine design and automation iema 2014 and computer science and application ccsa 2014

Tai ngay!!! Ban co the xoa dong chu nay!!! Industrial Engineering, Machine Design and Automation (IEMDA 2014) & Computer Science and Application (CCSA 2014) Proceedings of the 2014 Congress on IEMDA 2014 & Proceedings of the 2nd Congress on CCSA 2014 9622_9789814678995_tp.indd 3/3/15 5:07 pm May 2, 2013 14:6 BC: 8831 - Probability and Statistical Theory This page intentionally left blank PST˙ws Industrial Engineering, Machine Design and Automation (IEMDA 2014) & Computer Science and Application (CCSA 2014) Proceedings of the 2014 Congress on IEMDA 2014 & Proceedings of the 2nd Congress on CCSA 2014 Sanya, Hainan, China 12 – 14 Dec 2014 Editors Shihong Qin School of Electrical and Information Engineering Wuhan Institute of Technology, China Xiaolong Li College of Technology, Indiana State University, USA NEW JERSEY • LONDON 9622_9789814678995_tp.indd • SINGAPORE • BEIJING • SHANGHAI • HONG KONG • TA I P E I • CHENNAI 3/3/15 5:07 pm Published by British Library Cataloguing-in-Publication Data Conference  Proceedings  for  2014  Congress  on  Industrial Engineering, Machine  Design  and  Automation  (IEMDA 2014)  and  The  2nd  Congress  on  Computer  Science  and  Application  (CCSA 2014) system now known or to be invented, without written permission from the publisher Dr Ng - IEMDA 2014 and CCSA 2014.indd 17/2/2015 4:01:19 PM v Editorial It has been a great pleasure for me to welcome all of you to the joint conferences of 2014 Congress on Industrial Engineering, Machine Design and Automation (IEMDA2014) and the 2nd Congress on Computer Science and Application (CCSA2014), held in Sanya, China during December 12-14, 2014 During these two-days, international speakers presented their state-of-art research works in industrial engineering, machine design, automation, and computer science to solve today industrial problems We hope you enjoy this opportunity to share the results and make new connection for future collaboration The conference program consisted of two invited keynote presentations and invited sessions: Communication and Information Technology; Research and Design of Machines and Mechanisms for Manufacturing; Data, Signal and Image Processing, Computational Technology; Mechanical, Automation and Control Engineering This proceedings collected together the latest research results and applications on industrial engineering, machine design, automation, and computer science and other related Engineering topics All submitted papers were subjected to strict peer-reviewing by 2-4 expert referees, to ensure that all articles selected are of highest standard and are relevance to the conference On behalf of the organizing committee of IEMDA2014 and CCSA2014, I would like to take this opportunity to express our sincere appreciations and thanks to all authors for their contributions to the conference As well as to all the referees for their time in reviewing the articles and their constructive comments on the papers concerned Finally, as the editor of this proceedings, I am indebted to the support of the organizing committee for their hard works, World Scientific for their support in publishing this proceedings in such short space in time Without these vi excellent supports, the IEMDA2014 and CCSA2014 would not able to publish so timely and successfully Prof Shihong Qin Editor of IEMDA2014 and CCSA2014 vii Contents Editorial Chapter 1: Communication and Information Technology An Energy Consumption Assessment Method for WIFI Large-Scale Wireless Sensor Network Based on Dynamic Channel Energy Model W.K Tan, X.Y Lu, Y.X Xu, K.J Zhao and P Gao Research on Cloud-Based LBS and Location Privacy Protection Y Yan and W.J Wang Research and Exploration of the Hierarchical Management of Campus Network C.B Liu, T.Y Zhou, S.L Cai and K Sha 16 Improvement of Localization Algorithm for Wireless Sensor Network in Environmental Monitoring C Liu, S.L Wang, Y Ma and Z.Q Zhai 22 A New Study on Bank Interconnected System Security Solutions J Lin 29 Research on Simulation Platform for Security and Defense of Oil and Gas Network SCADA System Q.C Hu, X.D Cao, W.W Zhang, P Liang and Y Qin 34 Research on Print Concentrated Control Scheme Based on Virtual Print Technology Y Xie, X.L Zhou, Z.P Wen, G.L Li, S.P Liu and Q Hu 41 Developing Real Time SCM Information System C.T Huang, C.W Hsu, C.H Hung and W.L Wang 48 An Anomalous Traffic Detection System of the Controlled Network Based-On the Communication Rules W.J Han and Y Wang 55 viii The Study of HARQ Technology FDD-LTE Physical Layer Downlink W.B Tang and L.P Wang 61 Integrating with Information of Science and Technology on the Meta Data Ways’ Knowledge Base—the Pyramid Model to Aid the Decisions in Science and Technology K Hong, X Chen and J.R Hu 67 Sliding Window Frequent Items Detection in Wireless Sensor Networks S Wang and L.N Wu 76 Research on Security Mechanism of Cloud Security Immune System L Huo, J.X Zhou and X.W Liu 83 Design and Realization of Solar Cell Monitoring System Based on Wireless Sensor Network X.B Sun, Y Huang, J.J Tan, J.Q Yi and T Hu 90 Research of Multi-UAVs Communication Range Optimization Based on Improved Artificial Fish-Swarm Algorithm J.H Wu, J.Z Wang, Y.Q Cao, Y Cao and X.B Shi 98 Analysis of Limiting Factors and Numerical Simulation in Optical Transmission System B Yang and W.P Zhong 105 An Improved Timestamp-based User Authentication Scheme with Smart Card T.H Feng, C.H Ling and M.S Hwang 111 Cryptanalysis and Improvement of Zhuang-Chang-Wang-Zhu Password Authentication Scheme S.M Chen, C.S Pan and M.S Hwang 118 Chapter 2: Research and Design of Machines and Mechanisms for Manufacturing Research on Wind Power Simulation Model Y.S Zhang, A.N Tian and Y.L Pan 124 Model Test Study of Influence of Soil Bag Stacked Form on Ground Bearing Characteristics W Li, X.Y Shan and Z.B Jia 131 ix Analysis and Application of SMED in an Injection Molding Machine Based on VSM M Lv, N Zhang, J.C Jiang and Q Wang 143 Equivalent Mechanical Model to Support Real-Time Simulation of the Deformation of Thin-Walled Structures L.Z Tu, Q Yang, Y Zhuang, A.L Lu, Z Lin and D.L Wu 150 Nanoscale Sliding Contacts between Multi-Asperity Tips and Textured Surfaces: Effects of Indentation Depth T Zhang, D Wu, F Zhang, X.K Mu and R.T Tong 161 Based on Epsilon Method Structural Non-Probabilistic Reliability Analysis K Ma and H.P Fu 168 Research on Modeling and Simulation to Control WIP Inventory in EMS Enterprises Based on Bottleneck M Cai, T Shang, H.B Liu and H Chen 175 Screening Customer Order under Engineering-To-Order Environment H.E Tseng and S.C Lee 185 Genetic Algorithm with Unit Processor Applied in Fused Deposition Manufacturing (FDM) for Minimizing Non-Productive Tool-Path J Gong and L Zhang 191 Simulation and Analysis of Edge Cam Downward Mechanism Based on Contact Dynamics Model J Lu, J.J Zhang, F Lu and X.H Pan 198 Complex Product Collaborative Development Framework L.W Zhang and W Shen 206 The Similar Conditions and Similar Criterions of Deep-Sea Mining Experimental System Y Xu, X.F Zhang, L Liu and W.M Zhang 217 Reliability Analysis of an Air Supply System Design by Shortest Path Approach Based on Directed Network P Jiang and Y.Y Xing 224 Genetic Design of Integrated Manufacturing Supply Chains W Su, K.L Mak and B.L Qiu 230 483 Conclusion The dynamic output feedback H reliable controller with sensor failures is given by LMIs A numerical example has been provided to illustrate the effectiveness of the proposed design method Acknowledgment This work is supported by Natural Science Foundation of Liaoning Province, China (Grant 201202200) References F.Z.Wang, B.Yao, S.Y.Zhang, Guaranteed cost reliable control with actuator failures, Journal of Northeastern University, 7-24 (2003) 616-619 B.Yao, Z.J.An, F.Z.Wang, Robust and non-fragile H-infinity reliable control for uncertain systems with ellipse disk pole constraints, Proceeding of the 11th World Congress on Intelligent Control and Automation, (2014) 3964-3696 H.Hu, B.Jiang, H.Yang, P.Shi, Non-fragile Reliable H ∞ Control for Delta Operator Switched Systems, Proceeding of the 11th World Congress on Intelligent Control and Automation, 1(2014) 2180-2185 M Meenakshi, M S and Bhat, Fixed order robust H stability augmentation for Micro air Vehicle Design and validation, IEEE International Conference, (2013)1–6 L.Yu, Robust Control—LMI approach Tsinghua University, China, 2002, pp 59-60 G.B.Cai, and B.J Yin,X.J.Han,C.H.Hu, and H.F He, Dynamic Output Feedback H Control for Continuous-time Polytopic LPV Systems, Proceedings of the 33th Chinese Control Conference, 1(2014)3603-3608 Y.C Wu, L.F Ma, Y.M Bo, and X.H Liu, Reliable Mixed H / H ∞ Control for Stochastic Time-Varying Systems Against Actuator Failures, Proceedings of the 33rd Chinese Control Conference, 1(2014)3059-3064 484 Application Research of Neural Network Hybrid Modeling Method for Torque Measurement on Centrifuge Suspended Basket Trunnion ShengLai Chen† Institute of System Engineering, China Academy of Engineering Physics, Mianyang 621999 E-mail: †chensl@caep.ac.cn Because of the especial structure, it’s not feasible to instrument pressure transducer, or torque sensor The method of using strain gauge to measure the trunnion torque of centrifuge suspended basket is proposed Therefore, torque measurement system is imperative to be founded The mapping function of input and output is not absolute linear in practice A neural network (NN) hybrid modeling approach is proposed and applied to torque measurement system calibration The simulated studies on the calibration of single output system are conducted respectively by use of the developed hybrid modeling scheme The NN hybrid modeling approach is utilized to calibrate torque measurement system prototype based on the measured data obtained from calibration tests The simulated and experimental results show that the NN hybrid modeling approach can improve significantly calibration precision in comparison with traditional calibration methods In addition, the NN hybrid modeling is superior to NN black box modeling because the former possesses smaller network scale, higher convergence speed, higher calibration precision and better generalization performance Key words: Unbalanced moment, Calibration, Neural network, Hybrid modeling Introduction In industrial production, scientific research and daily life, torque measurement requirement is very extensive The rotation shaft torque test is the most common Either dynamometer or small torque wrench, different kinds of torque measurement device can be bought from market The schemes to solve rotation shaft torque measurement are not identical, such as the traditional electromagnetic or strain type, the new method of novel surface acoustic wave or inverse magnetostrictive method [1-4] Centrifuge is a kind of test equipment used for acceleration test The test piece is installed for uniform circular motion in centrifuge test, and the centripetal acceleration is used to simulate the acceleration environment use, that is the basic principle of the centrifugal test The centrifuge arm shelf structure is shown in Fig.1 Through the fixture, the test piece is arranged on the centrifuge basket, which is an important part of 485 fixture and test pieces, usually made of steel, a shaft trunnion center symmetric structure, and is connected with the centrifuge trunnion through both sides of the rotary arm The acceleration direction is from centrifuge test pieces centroid to the axis of rotation The basket can rotate around the shaft, to change the relative attitude of test piece, and change the direction of the acceleration test Then centrifuge basket must be locked through the locking device, which usually is expanding sleeve or wedge locking structure, in order to avoid the basket rotating around the shaft during centrifuge operation process Usually, there is the maximum unbalanced torque index for locking device Therefore, counterweight block must be installed rightly on the other side of centrifuge basket before the test to balance the moment of inertia test piece, it is called balancing the centrifuge basket For massive test pieces, often due to the size of the centroid position is not accurate enough (actual value and design value deviation or unable to accurately measure), in static or small overload condition, the torque trunnion borne is small, but in the case of high acceleration load, the torque will increase along with increasing of overload, and it may reach very great For example, if an 800kg test piece in the 20mm centroid position deviation, when the overload a reached 1200m/s2, unbalanced moment will reach 19200N • m For the kind of centrifugal test for container with un-full liquid, centrifugal basket trunnion imbalance torque is changing along with acceleration value, and the unbalanced torque maybe reach great[5] Therefore, it is very important to establish online monitoring system of unbalance torque, especially for the kinds of centrifugal test such as massive test pieces, high overload, and container test pieces with un-full liquid Fig Centrifuge arm shelf structure 486 Analysis of Measurement Principle Strain torque measurement is a conventional method of torque measurement The strain of the rotating shaft surface is measured, and then the value will replaced into corresponding mechanics formula, finally the torque can be obtained Centrifuge basket shaft deformation is mainly caused by the bending moment and torque, the strain got from measurement system should be produced only by the torque Four strain gauges, whose resistance value and the sensitivity are equal, are chosen to form bridge arms, and the angle between strain gauge and the shaft surface circumference should be ±45°, as shown in Fig 2(a) Where, R1, R2, R3 and R4 respectively represent the resistance of strain gauge There is R1=R2=R3=R4, group of the bridge is shown in Fig 2(b) (a) (b) Fig Schematic arrangement of strain gauge With the external force affection, strain gauge reformate together with the shaft at the measuring points, deformation, denoted as ε Mx and ε My , and when the temperature changes, temperature strain gauge, denoted as ε t , so the strain of gauges can be expressed as ε1 = −ε Mx + ε 1My + ε t1  2 ε = ε Mx + ε My + ε t  3 ε = −ε Mx + ε My + ε t  4 ε = ε Mx + ε My + ε t Because the resistance R, the sensitivity coefficient K and temperature t are the same, there is ε t1 = ε t2 = ε t3 = ε t4 Strain gauge and strain gauge 4, strain 487 gauge and strain gauge respectively in the same bus, so ε 1My = ε My , ε My = ε My The strain slice groups of full bridge, strain gauge reading ε = ε1 − ε + ε − ε = 4ε Mx , that is ε Mx = ε / Visibly, the strain obtained from full bridge measurement completely due to torsion deformation Torque measurement system of equations is[7]: Mx = W Eε Mx WE ε = ⋅ (1 + µ ) (1 + µ ) where W is the section modulus in torsion, E is the elastic modulus of materials, G is the shear modulus of materials, and µ is the Poisson's ratio of the material Visibly, the relationship between torque Mx and measuring point strain is linear, and can be expressed as: Mx = k ⋅ ε Through measuring the output strain under different torque, to determine the coefficient K is called calibration of torque measurement system In practice, because of the influence of structure design, processing technology, the location and angle error and other factors, it is not completely satisfy the linear relation the system mapping of input and output can be simulated by hybrid neural network modeling method Hybrid NN Modeling Method For a complex nonlinear system, in most circumstances, one can always get an original model based on a priori physical or empirical knowledge by use of the first principle computations and conventional parameter identification methods at the earlier stage of modeling This model is also referred to as the knowledge-based model which may be coarse but commonly reflects the primary characteristics of the system For the unknown and un-modeled nonlinear characteristics of the system, one can employ non-parametrical identification methods such as the NN approach Once network training is successful, the acquired NN model is integrated with the knowledge-based model to yield a hybrid NN model which is able to describe the entire system property The above procedure is termed the hybrid NN modeling methodology In this approach, the NN model provides a complement to the knowledge-based model so as to reduce the modeling errors Three kinds of topological architectures have been applied in the hybrid NN modeling: serial, parallel and built-in configurations In the serial configuration, the knowledge-based models provide the inputs for the NN models; conversely, the outputs of the NN models 488 can be used as the inputs of the knowledge-based models In the parallel configuration, the knowledge-based models and NN models represent parallel subsystems In the case of built-in configuration, the NN models are embedded into the knowledge-based models[8] or the latter is embedded into the former[9] Generally, the unknown and un-modeled nonlinear characteristics are ‘weak’ and ‘local’ in comparison with the primary characteristics of the system (described by the knowledge-based model) It leads to the fact that the hybrid NN modeling strategy can significantly improve network training efficiency and training precision Moreover, due to the inclusion of the knowledge-based model, the hybrid NN model is more apparent and possesses better generalization performance than the pure black-box NN model[8-11] The paper introduces the hybrid NN modeling scheme into the modeling of torque measurement system with complex nonlinear input-output properties Calibration of the neural network modeling method is applied to the test of hybrid system, function relation between input and output of system can be expressed as: Mx = f (ε ) In this formula, Mx represents the physical quantity to be measured, ɛ is the output signal When the function is linear, simple form or structure is known, the calibration results can obtain better precision by fitting the measured data samples But when the structural form is unknown and complex, and the function is nonlinear, the traditional fitting method by assuming different model forms and the measurement data is repeated fitting to complete the calibration, not only time-consuming and laborious, the precision cannot be ensure The function can be rewritten as Mx = f1 (ε ) + f (ε ) where f1(ɛ) is obtained according to the characteristics of sensor calibration and measurement variation data, assuming that the model structure is appropriate, and data fitting It can describe the main characteristics of the measured function between the physical quantity and the system output signal, called knowledge base model f2(ɛ) represents the error between knowledge base model and actual model, and it is weak, local form of unknown, complex and difficult to parametric modeling It can be obtained by the modeling method of neural network In this way, f1 and f2 form a parallel neural network hybrid model, which can describe the measured actual variation of physical quantity and the sensor output signal effectively On the other hand, can also according to the sample data, only using the neural network model directly on the function 489 relationship between the measured and the sensor output signal, namely the pure black box modeling Two modeling methods are compared in the following Calibration Test According to the torque testing scheme, a full bridge strain gauge torque test The subject is a rotating shaft of interference fit shown in Fig.3, the structure is described in detail in the literature[12] Full bridge strain gages are pasted at 45° angle with the axis, as shown in Fig.3 The selected different position is arranged at points, according to the measurement results the most reasonable point is chosen for monitoring test Test loading device as shown in Fig.3, synchronous loading using static loading system, two hydraulic cylinders are respectively applied vertical pulling force and pressure The arm length is 0.5m Strain measurement is in progress of 6kN • m calibration test and 8kN• m monitoring test continuously The largest strain values measured data is chosen as calibration data The results of linear calibration method and neural network hybrid modeling method, are shown in Fig.4 comparatively Among them, the black box component neural network hybrid model is BP network structure of 20 hidden layer nodes The network parameters set as follow: net.trainParam.goal=0.02, hidden layer transfer function is tansig, net.trainParam.epochs=2000, the output layer transfer function is purelin, training function is trainlm After the 694 steps it meet the requirement According to the model obtained by calibration methods, strain measuring data input 8kN • m monitoring test simulation results are compared with the experimental data, the results are shown in Fig.5 Obviously, the method of neural network hybrid model method is better than the linear Fig.3 Calibration test 490 Torque/kN•m Error of linear calibration Torque/kN•m Error of hybrid modeling linear calibration hybrid modeling Strain/10-6 accuracy value Strain/10-6 Fig.4 Results and error of different calibration methods Error of linear method output Torque/kN•m Output of linear method Torque/kN•m Error of hybrid modeling output Output of hybrid modeling accuracy value Strain/10-6 Strain/10-6 Fig.5 Results and error of monitoring test by different calibration methods Conclusion Calibration of measuring system is an important link in measurement technology, and it influence precision and reliability of test results directly For the nonlinear characteristics system, the map relationship between the measured physical quantity and output signal is complex, unknown and nonlinear The linear calibration method is difficult to achieve satisfactory accuracy Using the neural network hybrid modeling method is effective, and neural network hybrid modeling methods and procedures are given The simulation research and actual calibration results show that, compared with the traditional linear calibration method, hybrid neural network modeling method can effectively compensate the calibration error leaded by the sensor input and output complex nonlinear characteristics, and significantly improve the calibration precision sensor On the other hand, compared with black box, 491 hybrid neural network modeling method has such superiority as small network scale, fast convergence, high accuracy and generalization ability References Guanlin Shen The new development of strain electrical measuring and sensor technology and application [J] Chinese test, 2011, (2): 87-96 Lixia Liu Vehicle torque and rotational speed test system[J] Instrument technique and sensor, 2010, (7): 89-91 Yan Wang, Jiangwei Zhu Present situation and development trend of torque measuring methods[J] Journal of woodworking equipment, 2010, 38 (11): 14-18 Shihong Xie, Haixiong Nie, Xiaogang Wang Research of spindle torque plate straightening machine in the on-line monitoring system[J] Journal of heavy machinery, 2011, (1): 39-42 Shenglai Chen, Changchun Zhu A tank of hanging basket centrifuge test of dynamic balance analysis [J] Journal of environmental engineering equipment, 2010, (6) : 243-246 Dengquan Wang, Ming Yang, Lin Ye Non-contact rotation torque measurement status[J] Journal of electronic measurement technology, 2010 (6): 8-12 Chen Shenglai, liu qian Centrifuge basket trunnion torque testing technology[J] Journal of environmental engineering equipment, 2013, 10 (1): 102-104 M Cao and K.W Wang: A hybrid neural network approach for the development of friction component dynamic model J Dyn Syst Meas Control, 126 (2004), 144-53 Y Oussar and G Dreyfus: How to be a gray box: dynamic semi-physical modeling Neural Netw, 14 (2001), 1161-72 10 M Cao, K.W Wang and W.E Tobler: Advanced hybrid neural network and automotive friction component model for powertrain system dynamic analysis: part Model development Proc Inst Mech Eng D, 218 (2004), 831-43 11 C.G Zhou, X.N Zhang, S.L Xie: Hybrid Modeling of Wire Cable Vibration Isolation System Through Neural Network, Math Comput Simul, 79 (2009), 3160-3173 12 Qian Liu, Zhigang Deng Centrifuge basket design of locking mechanism study[J] Journal of environmental engineering equipment, 2010, (6): 241-243 492 Study on Additional Damping Control Strategy of Permanent Magnet Synchronous Generator Shengyong Ye1,a, Yitian Zhang2,b, Renjun Ruan1, Quan Tang1, Songling Dai1 and Tianyu Wang2 Sichuan Electric Power Corporation Economic Research Institute, Chengdu, China, 610041 Southwest Jiaotong University, Chengdu, China, 610031 E-mail: ayeshengyong410@sohu.com, bzhang_yitian312@163.com This paper presents an additional damping control strategy through adjusting the rotor speed of the direct-driven permanent magnet synchronous generator (D-PMSG) The parameters of the controller are optimized by PSO method The simulations are done in a modified machine zone system, considering both constant and variable wind speeds The results show that the design of the additional damping control strategy can significantly enhance the system damping and inhibit inter-area low frequency oscillation Keywords: Direct-driven Permanent Magnet Synchronous Generator; Additional Damping Control Strategy; Low Frequency Oscillation Introduction Vigorous development of wind power is an important development direction of future energy strategy of China A number of large-scale wind power bases are being built currently in our country However, inter-area low frequency oscillation of interconnected power grid has become an important factor restricting power transmission capability and large-scale wind power integration 1-2 So, putting forward feasible control strategies of power system low frequency oscillation, and improving power system security and stability as well as the safe and stable operation of wind turbines has important significance In order to inhibit low frequency oscillation, the existing literature show that we can enhance system damping by using first-system strategy such as energy storage device3, enhancing grid4 and so on, but the cost is higher By contrast, second-system strategies, such as additional damping control have the 493 advantages of low cost and easily implementing the project The literature1 puts forward the concept of dynamic frequency characteristics of wind farm, and the design method of an additional damping control strategy of new doubly fed wind turbine The reference5 brings the doubly fed motor slip signals in the rotor side converter control model, and improves the system damping through changing the phase output damping power of the rotor excitation voltage The reference6 puts forward an improved power control method to inhibit the system low frequency oscillation and improve the stability of the system The reference designs permanent magnet direct-driven wind turbine additional damping controller by introducing the torque coefficient method The reference8 stores the unbalanced energy of system oscillation by using wind turbine speed change Because permanent magnet direct-driven wind turbines can regulate speed, shafting inertia is large, and it has flywheel energy storage effect Also, for permanent magnet direct-driven wind turbine suite, output power is controlled by a converter, which has the advantages of being fast and controllable Additional damping control does not change the original control mode of wind turbine; it still uses the maximum power point tracking under rated wind speed, making the maximum use of wind energy It improves the utilization efficiency of wind energy and enhances system damping at the same time, which providing a good solution to the lack of system damping problem of regional power grid with high permeability wind power Also, it can improve grid stability and ensure the safe and stable operation of the power grid and the wind farm Wind Turbine Model and Control Strategy A simple schematic diagram of wind energy generation system with direct-driven permanent magnetic synchronous generator (PMSG) is shown in Fig us is U dc ig ug ω ω v Ps _ ref Qs _ ref U dc _ ref Qg _ ref Fig Schematic diagram of D-PMSG and converter control system 494 2.1 Aerodynamic model According to the aerodynamics theory, the mechanical power of the wind turbine is expressed by the function:   Pm = ρπ R v C p ( λ , β )  λ = Rωr  v (1) Where ρ is air density; πR2 expresses the rotor swept area; ν is wind speed; Cp(λ,β), the power coefficient of wind turbine, is the function of λ and β It expresses the efficiency of wind utilization β is the pitch angle λ is the tip-speed ratio, which means the ratio of tangential speed of the blade tip to the wind speed 2.2 PMSG model The voltage equations of the PMSG are described in the d-q coordinate: diqs  − pωr Lds ids + pωrψ f uqs = − Rs iqs − Lqs dt  dids  + pωr Lqs iqs uds = − Rs ids − Lds dt   Te = p ψ f iqs + ( Lds − Lqs ) iqs ids   (2) Where uds, uqs, ids, iqs are the stator voltages and currents; Lds, Lqs are inductances of the direct and quadrature axis; Rs is the resistance of stator; ψf is permanent magnet rotor flux 2.3 Converter control model The grid-side converter maintains a constant DC voltage by adjusting the d-axis and q-axis current, to achieve active power and reactive power decoupling control and control the flow of reactive power to grid The grid-side converter usually runs in the unity power factor operation status The equation of grid-side converter is as followed: 495  x
4 = VDCref − VDC   x
5 = K p (VDCref − VDC ) + K i ∗ x4 − idg   x
6 = iqgref − iqg  vdg = K p ∗ K p (VDCref − VDC ) − K p ∗   K i ∗ x4 − K p ∗ idg − Ki ∗ x5 + iqg X net (3) Where Kp4, Kp5, Ki4, Ki5 are the proportional and integral constant of PI controller; x4, x5, x6are the intermediate variables; vdg and vqg are d-axis and q-axis components of the grid side; idgref and iqgref are the d-axis and q-axis reference components of the stator current’s; ω, ωref is the wind turbine angular velocity and its reference value The rotor-side converter controls the active and reactive power of generator by adjusting the d-axis and q-axis current to enable the generator operate on the maximum power pointing tracking under the rated wind speed The equation of generator converter is as followed:  x
1 = idsref − ids   x
2 = ωref − ω   x
3 = K p (ωref − ω ) + K i ∗ x2 − iqs  vds = K p1 ( idsref − ids ) + K i1 ∗ x1 + iqsωs Ld  vqs = K p ∗ K p (ωref − ω ) + K p ∗ K i ∗  x2 − K p ∗ iqs + Ki ∗ x3 − idsωs Ld  (4) Where Kp1, Kp2, Kp3, Ki1, Ki2, Ki3 are the proportional and integral constant of the PI controller; x1, x2, x3 are the intermediate variables; vdg and vqg are d-axis and q-axis components of the grid side; idgref and iqgref are the d-axis and q-axis reference components of the stator current; ω, ωref is the wind turbine angular velocity and its reference value 2.4 PMSG additional damping controller PMSG is decoupled from power system through full power converter, without reaction to system oscillation, and thus can't provide damping support According to the theoretical derivation in section 2, adding the feedback signal of system oscillation to PMSG active power control loop can effectively improve the system damping, and eliminate the system power oscillation When system oscillation occurs, the additional damping controller adjusts the reference 496 of wind rotor angular velocity reference after responding to oscillation The permanent magnet synchronous generator can eliminate power oscillation through injecting damping power related to system oscillation damping For permanent magnet direct-drive wind generators, wind turbines that control output power through inverters, have the advantage of rapidity and controllability When the generator rotor slows down, the output power increases; conversely, output power decreases Additional damping control does not change the original control way of wind turbines When the wind speed is below the rated value, wind turbines always operate on the maximum power pointing tracking Consequently, wind turbines can make the best use of wind energy and improve the utilization efficiency of wind energy ωdamp + ωref + K p2 + −  + sTa     + sTb  K K i iqsref + s K p3 + − iqs ω n sTw + sTw Pac + Vqs Ki3 s − ωs Ld ids ωs Ld − idsref + − K + K p1 + i1 s Fig The rotor side converter control + Vds The additional damping controller, as shown in the dashed part of Fig 2, consists of the DC blocking link, the phase compensation and the gain The DC blocking link is a differential link, whose main function is that the output is zero at steady state, and that the output is the dynamic signal at system oscillation, so that the additional damping control only works when the system oscillation occurs The phase compensation link is composed of a lead-lag link, and provides phase compensation, so that the additional moment ∆Te has the same phase as the angular velocity ∆ω The amplification link ensures enough amplitude of the additional torque amplitude The additional damping controller transfer function G(s) as follows: G (s) = KsTw (1 + Ta ) n (1 + sTw )(1 + Tb ) n (5) 497 Where K is the gain; Tw is the time constant of the DC blocking link; Ta and Tb are the time constant of the phase compensation, n is the number of phase compensation, and generally takes value in 1~3 2.5 Parameter setting method The additional damping controller parameters are optimized by particle swarm optimization (PSO) The basic idea of particle swarm optimization is that some particles without volume and mass are initialized stochastically, and that each particle is regarded as a feasible solution of the optimization problem Through setting the objective function, the iteration is available to seek the optimal solution of the particles In the iteration, the particles will track two extreme values: the one is the optimal solution that the particles themselves have found so far (individual extreme pbest) and the other one is the optimal solution have found so far (group extreme gbest) The PSO algorithm has a simple concept, and less control parameters The optimal result has nothing to with the initial value The PSO algorithm parallels to some extent In the iteration, the speed and position of each particle is updated by the following equation vk +1 = wvk + c1 × rand × ( pbestk − xk ) + c2 × rand × ( gbestk − xk )   xk +1 = xk + vk +1 (6) Where, vk is the velocity vector of particles; w is inertia weight factor; c1and c2 are learning factors; rand is the random number between and 1; xk is the position vector of particles Too small population size will cause substantial increase of iteration number, and too large population size will cause the decrease of search efficiency The population size is taken generally between 30 and 50 Finally the optimal parameters of the controller are determined as Ta=0.13, Tb=0.46, K=0.03 Simulation and Analysis 3.1 Simulation model ， As shown in Fig a four-generator two-area system (system parameters refer to Reference 15) connects with a permanent magnet direct drive wind farm The results of power flow calculation show that the wind power transmission lines can meet the requirements for wind farm to send out power There is no line power flow in the network exceeds the limit, and all the bus voltages are in the reasonable range of deviation