| With the large-scale application of street lighting facilities and the continuous development of smart cities in China,the routine maintenance and troubleshooting of street lighting is the key to the normal operation of urban lighting.At present,the majority of street lighting fault diagnosis is still carried out using traditional methods,and there are limitations to the troubleshooting and maintenance guidance;and in terms of street lighting maintenance inspection,there is a lack of predictability for areas where faults may occur,relying more on human subjective predictions.With the widespread use of urban intelligent lighting systems,research and analysis of common faults in street lighting has revealed that there is a degree of correlation between street lighting faults and the street lighting operation data collected by the lighting system.Therefore,in the context of the Guilin Key R&D Project "Road Smart Lighting and Operation and Maintenance System Based on Big Data Artificial Intelligence Technology"(Project No.2019021113),this paper investigates a data-driven approach to streetlight fault diagnosis and early warning,aiming to achieve fault diagnosis for specific categories of streetlights,and to provide a better understanding of future operational trends and status of streetlights for prediction and early warning.The main research work of the thesis is as follows.(1)An improved VMD(Variational Mode Decomposition)and feature selection method for street light signal analysis and processing is proposed,using the faulty street light operation data collected from the urban intelligent lighting system in Chongzuo,Guangxi Zhuang Autonomous Region as the experimental data source,and mining the correlation information between specific fault categories and operation data.Firstly,in order to reduce redundancy and noise,the street light operation data were pre-processed using principal component analysis,and the main variable parameters were selected as input parameters for subsequent fault diagnosis and early warning.Secondly,signal processing method feasibility experiments were conducted to select variable modal decomposition for signal processing of the input parameters,so as to decompose them into modal components containing different scales of fault characteristic information.At the same time,to address the problem that the artificially selected VMD decomposition parameters may cause poor signal decomposition,this paper introduces the whale optimisation algorithm to improve the VMD decomposition parameters and achieve the adaptive optimisation of the best decomposition parameter combination.The experimental results of the VMD fault signal analysis based on the improved whale optimization algorithm show that there are obvious periodicity and trend in the current and historical time windows of the operating signals of the variable parameters of different fault categories.Further,in terms of fault feature selection,this paper introduces Pearson correlation coefficient and cliffness to filter the modal components with high correlation with the original signal,and uses the sample entropy and cliffness values as street light fault feature indicators to construct a fault feature vector set.Through experimental analysis,it is proved that the method in this paper has excellent fault feature extraction effect.(2)In terms of building fault diagnosis models,this paper proposes an improved XGBoost-based fault diagnosis model for street lighting,and the model is trained on the fault feature vector set constructed by the improved VMD signal analysis and fault feature preference method.The model comparison experimental results show that the fault diagnosis model using XGBoost has better diagnosis effect and feasibility compared with KNN,decision tree and BP neural network algorithms.Further,to address the problem that the hyperparameters of the original XGBoost model tend to affect the diagnostic accuracy,this paper introduces the sparrow optimization algorithm to optimize the hyperparameters of the XGBoost model,so as to enhance the generalization ability of the model.The experimental results show that the improved XGBoost model can improve the model fault diagnosis accuracy from 91.6% to 96.87%,and the diagnosis accuracy is significantly improved.(3)An effective trend prediction and state warning mechanism for street light operation is established,and a trend prediction and state warning model based on Bi LSTM(Bidirectional Long Short-Term Memory)is proposed.Firstly,the LSTM network is used to predict the time series trend of the variable parameters of street light operation data.Secondly,to address the problem of poor prediction performance of LSTM networks,Bi-LSTM,a bi-directional time-series network,is used to improve the model.Experiments are conducted with real street light failure cases to verify the prediction effectiveness of the proposed model.Meanwhile,the comparison experimental results with other prediction methods prove that the Bi-LSTM model can overcome the shortcomings such as cumulative error and poor prediction effect to achieve effective prediction.Based on the research of time series trend prediction of street light operation parameters,this paper adopts a time-sliding window combined with the statistical Lajda criterion as a state warning method to reasonably determine the threshold value for abnormal street light operation state,thus avoiding the shortcomings of using a single threshold warning mechanism.Finally,the experimental analysis results of real fault cases show that the proposed early warning method can detect the signs of faults in the abnormal operation of street lights in advance,thus realising early warning of faults in street lights,which has certain theoretical significance and application value. |
PDF Full Text Request