| At present,the construction of the fifth-generation mobile communication system(5G)is progressing steadily,which not only brings user experience of large bandwidth,low delay and large-scale connection,but also provides technical support for industrial upgrading.China is actively promoting the integration of 5G and new power systems.With the rapid development of the energy Internet and the construction of new power systems,the structure of power grid is more and more diversified and complicated.Moreover,emerging businesses such as distributed new energy,electric vehicle charging piles,intelligent patrol and accurate load control keep coming up,the number of power terminals and the amount of data increase rapidly,and the delay sensitivity of control businesses of new power systems increases significantly.The existing communication approach in current power system with sub-6 GHz frequency band is unable to meet the demands of ultra-dense connection,super-large bandwidth and ultra-low delay of future electric power service.In order to increase the communication rate and system capacity,millimeter wave(mmWave)communication comes into being.MmWave will be used as the"golden band" in the next generation wireless communication system for its continuously and available bandwidth,as well as acceptable signal propagation distance.MmWave has become one of the key technologies in the evolution of 5G and the sixth generation of mobile communication,which also has an expectant application prospect in the new power system.It is necessary to accurately know the transmission characteristics of wireless signals before developing,designing and constructing a new wireless communication system,and the measurement and accurate modeling of wireless channels is an important premise to clarify the law of wireless signal transmission.The high attenuation,sparsity and time-varying characteristics of mmWave signal make the channel characteristics significantly different from that of sub-6 GHz frequency band,thus existing methods cannot be applied in accurate channel modeling and simulation of mmWave in different scenarios.Therefore,mmWave channel measurement and accurate time-varying channel modeling and simulation is of great significance for understanding wireless signal propagation,designing and constructing mmWave communication systems.This paper conducts research on mmWave channel modeling and simulation approach for the power distribution and utilization scenarios.Based on measured mmWave channel data,this paper focuses on exploring more accurate and efficient mmWave time-varying channel modeling and simulation methods by introducing artificial neural networks,deep learning and other approaches,making up for the shortcomings of traditional channel modeling and simulation methods.The main work and innovations of this paper are as follows:(1)MmWave time-varying channel measurement in power distribution and utilization scenarios,and relevant scenario and state identifications approachIn order to make up for the blank of channel parameters and models of power distribution and utilization scenarios in existing standardized documents,this paper carried out mmWave channel measurements in the substation,the waiting hall of high-speed railway station,etc.,the channel characteristics in multiple scenarios and states are analyzed and listed in parameter tables.Aiming at the problem that the traditional channel identification method can only identify scenario or state,and the accuracy is relevantly low,this paper proposes a deep neural network(DNN)based mmWave scenario and state identification approach based on the actual mmWave time-varying channel measurement data.The DNN structure is designed and the relevant superparameter configuration is determined,which enables it to identify both channel scenario and state.Different from the existing methods which only consider the statistical channel features,the DNNbased approach introduces the cluster-based channel features,which bring environment information.The proposed method is verified by the measured channel data,the identification accuracy of the proposed DNN-based method is higher than the existing methods,the identification accuracy of channel scenario,state and overall are 99.83%,98.85%and 98.68%.(2)Semi-deterministic mmWave time-varying channel modeling and simulation approachIn order to solve the problems that traditional statistical channel model is difficult to conduct link-level simulation,deterministic channel model requires detailed environmental information and large amount of computation,and existing neural network based approaches are difficult to achieve accurate parameter prediction in the case of small data,this paper proposes a semideterministic mmWave time-varying channel modeling and simulation approach based on optimized neural network(ONN)to realize the accurate channel simulation in a custom environment.Based on the proposed improved simulated annealing algorithm,the optimal parameter configuration of radial basis function neural network is searched to reduce the prediction error,and based on the optimal parameter configuration,the ONN-based parameter prediction model was constructed.The scatterers in actual environment are abstracted into determined and random generation ones,and the calculation methods of corresponding cluster parameters,multipath parameters and channel impulse response matrix are derived.The ONN is applied in the time evolution simulation to accurately predict channel parameters and ensure the continuity of channel simulation.The results show that compared with the existing methods,the ONN generated by the improved simulated annealing algorithm can achieve accurate channel parameter prediction in the case of small data.The proposed approach can achieve accurate channel simulation under the condition of customized transceiver layout,and the simulation accuracy of channel impulse response matrix can be improved by more than 40%.(3)Hybrid mmWave time-varying channel modeling and simulation approachIn order to solve the problems that the existing channel parameter prediction methods are not suitable for mobile applications,and the existing simulation methods are difficult to achieve accurate channel simulation beyond the measurement range under the condition of small data,this paper proposes a hybrid mmWave time-varying channel modeling and simulation approach based on deep learning,and realizes the combination of DNN,ray tracing and statistical channel modeling.Based on long short-term memory network and joint neural network,two prediction models are proposed to predict the channel parameters respectively for fixed and mobile applications.Moreover,based on the parameter prediction model,the statistical model of intra-cluster small scale channel parameter and the simplified environment model,the hybrid mmWave timevarying channel simulation approach is proposed,the calculation methods of the large-scale parameters,multi-path parameters and channel impulse response are derived.The results show that compared with the existing methods,the proposed approach can achieve accurate parameter prediction beyond the measurement range in the case of small data.The proposed method can guarantee the accuracy of channel simulation while greatly reducing the amount of ray tracing.The simulated multipath parameters and channel impulse response are in good agreement with the measured ones. |
PDF Full Text Request