Research On Key Technologies Of Equivalent Modeling,Monitoring And Control For Wind Farm Based On Multi-Type Learning

Wind power generation is one of the important new energies with the advantages of being clean,environmentally friendly and renewable.Under the guidance of the “dual carbon”strategic goal,wind power has been encountering the historical opportunity of rapid large-scale construction and high-quality development.With the continuous improvement of wind power penetration,the wind farm(WF)should not only meet their own security and stability requirements,but also become the participants in high-quality operation of power grid gradually.Especially,the WF should play an active defense and support role when there is a fault in the power system.In the event of a system fault,on the one hand,the WF needs to coordinate the action behaviors of dozens or even hundreds of wind turbines(WTs)in a very short period,so as to utilize their support capabilities;on the other hand,if the fault occurs in the WF,it is necessary to identify the reason of the fault quickly,remove the WTs affected by the fault in time,isolate the fault safely,and use the grid-connected WTs to assist the system to normal operation flexibly.With the continuous expansion of the WF scale,it is more and more difficult to achieve the above goals accurately,quickly and reliably.In recent years,multi-type learning techniques in the field of artificial intelligence have developed rapidly and achieved better performance than traditional methods in some fields,providing a chance to solve the above problems.Thus,this paper focuses on the research on key technologies of equivalent modeling,monitoring and control for wind farm based on multi-type learning.Firstly,this paper establishes the equivalent model of WF;then,the methods of WF fault monitoring and fault isolation control,WF fault ride through control and WF frequency recovery control are proposed;based on these studies,this paper presents the combined WF fault monitoring,fault ride through and frequency recovery control process(W4FP)application framework considering the energy storage;also,the corresponding intelligent solution methods based on multi-type learning are proposed according to the characteristics of the research problems.The main research contents include:(1)Aiming at the limitations that the detailed WF models are hard to achieve fast calculation and accurate convergence,and the existing WF equivalent methods perform poorly in terms of simulation view and multi-scenario applicability,the WF equivalent modeling techniques based on multi-view incremental transfer learning are proposed.Firstly,a “pointline-surface-space” representation method is proposed to describe the operating characteristics of WT,and the grouping index for WTs is constructed.On the basis of multi-view fuzzy C means clustering(MV-FCMC),multi-view transfer fuzzy C means clustering(MVT-FCMC)and multi-view incremental transfer fuzzy C means clustering(MVIT-FCMC)are proposed to achieve WTs grouping.Then,three equivalent models of WF based on MV-FCMC,MVTFCMC and MVIT-FCMC are proposed,respectively.Finally,the three equivalent models are simulated and verified,respectively;the multi-view simulation accuracy and multi-scenario applicability of the models are analyzed;the influences of WF scale and grouping index scale on the models are discussed.(2)Aiming at the requirements of single-phase grounding faulted line selection and fault isolation for the non-effectively grounded WF collectors,together considering the limitations that the WF fault monitoring models perform poorly in terms of transferability,and the fault isolation methods may cause large disturbances and threaten the safe operation of gridconnected WTs,the fault monitoring and fault isolation control strategy for the WF collector lines based on cross-domain feature adaption deep-transfer learning is proposed.Firstly,a data preprocessing technique for fault monitoring signals based on gradient similarity visualization is proposed,and the WF equivalent model is constructed based on MVIT-FCMC.Then,a cross domain feature distribution & dimension adaption based deep-transfer learning(C4DTL)algorithm is proposed.Taking the single-phase grounding faulted line selection of the collectors as the research scenario,a fault monitoring model based on C4 DTL for the WF collector lines is constructed.The model has good transferability and does not need to adjust its structure when WF topology changes,and only fine-tuning training for the parameter is required.Then,the paper puts forward a fault isolation control strategy considering the support of the WTs in the non-faulty lines,so as to ensure the smooth removal and isolation of the fault when the singlephase grounding occurs in the collector line.Finally,the accuracy,noise immunity and transferability of the fault monitoring model are analyzed;the effects of fault isolation control strategy on reducing disturbance and ensuring the safe operation of WFs are discussed.(3)Aiming at the circumstances that the system short circuit faults may cause WF to enter fault ride-through,and the problems that the existing WF fault ride through methods are difficult to achieve rapidity,accuracy and multi-scenario applicability meanwhile,a combined active/reactive power control strategy for WF fault ride through based on parallel deepreinforcement learning is proposed.Firstly,the model and overall framework of active/reactive power combined control in WF fault ride through are constructed.Then,the WT fault ride through model is established,and the method based on MV-FCMC is used to divide the WTs into different groups.Next,aiming at the shortcomings of the classical deep-reinforcement learning algorithms,a critic network free parallel based deep-reinforcement learning(CFPDRL)algorithm is proposed.Based on this,the control strategy based on CFP-DRL is presented,including the control model parameter design,model training method and model application method.Finally,the impacts of WTs divisions and deep-reinforcement learning algorithms on control are analyzed;the multi-scenario applicability,robustness and computational efficiency of the method are discussed.(4)Aiming at the possible power unbalance after fault ride through,and the existing WF frequency regulation methods may perform inefficiently and have difficulties in dealing with the complex control environment in changablely controllable WT resources,a WF frequency recovery control strategy based on hierarchical distributed hybrid deep-reinforcement learning is proposed.Firstly,the frequency recovery control model of the WT and the communication architecture in the control are constructed,and the hierarchical distributed control objectives in the frequency recovery control are defined.Then,the structure of the communication network is considered,and the WTs are divided into different groups using the MVT-FCMC based method.Next,according to the characteristics of the frequency recovery control model,a discrete consensus based hierarchical distributed hybrid deep-reinforcement learning(DCHDH-DRL)algorithm is designed.On this basis,considering the changablely controllable WT resources after the fault clearness,the communication rules in the control are designed,and the the details of the DC-HDH-DRL based controller are presented.Finally,this paper analyzes the control process of the model when all WTs are grid-connected and some WTs are cut off;the influences of clustering methods and deep-reinforcement learning algorithms on the control are discussed;the multi-scenario applicability and performance under abnormal communication condition of the model are analyzed.(5)Under the background of the development trend and demand of wind-storage coordinated operation,the application framework of W4 FP is proposed,and the application framework and supporting role of energy storage in W4 FP are presented.Firstly,the methods of WF equivalent modeling,fault monitoring,fault ride-through control and frequency recovery control are coordinated to construct the W4 FP application framework.Then,the features of various types of energy storages and their coordinated operation with wind power are analyzed,and adiabatic compressed air energy storage(A-CAES)is selected as the research object.Based on this,the supporting role of A-CAES in W4 FP is determined,and the model of A-CAES is established.Next,the W4 FP application framework considering A-CAES is designed.Finally,the W4 FP is simulated and tested;the influences of various factors on W4 FP are discussed;the W4 FP considering A-CAES is analyzed.