Research On Identification Method For Electric Bicycle Charging Load Based On Non-intrusive Load Monitoring

In recent years,electric bicycles have gradually become the main means of transportation for urban residents to travel short distances.The problem followed is that the number of fire accidents caused by electric bicycle charging has also shown a rapid growth trend.In order to effectively control fires accidents caused by illegal parking and charging of electric bicycles,the Ministry of Public Security and many provinces have successively implemented relevant regulations to regulate the parking and charging behavior of electric bicycles,and strictly prohibit illegal parking and charging behaviors.However,it is difficult and inefficient to supervise it through manpower.With the rapid development of load monitoring technology,non-intrusive load monitoring method provides a new solution for the supervision of illegal charging behavior of electric bicycles.There are many models of electric bicycles,with low power and unobvious features,resulting in low load identification accuracy.Based on the technical framework of non-intrusive load monitoring,this paper makes improvements and innovations in key technologies such as feature selection,event detection,feature extraction and load identification algorithms,and verifies them through examples of measured data.The main research process of this article is as follows:Firstly,this paper analyzes the charging load current characteristics of electric bicycles from the components and working principles.The resident load data collection platform is built to collect the voltage and current data of home appliances synchronously.15 load features,including PQ features,current features,V-I trajectory features,harmonic features and so on,are quantified from the perspectives of time domain,frequency domain and trajectory.According to the measured data,the load features of typical electric bicycles charging and household appliances are compared.Secondly,in order to filter out a subset of load identification features with high discrimination and low redundancy,a semi-supervised Fisher scoring method is proposed to fully mine the information of the semi-supervised sample set and measure the discrimination of load features.The maximum information coefficient is used to measure the redundancy between load features.A multi-objective optimization model for load feature selection is established,and the model is solved by a greedy algorithm.Research has found that feature selection removes irrelevant or redundant features,improves the recognition effect,and reduces the feature dimension at the same time.Then,considering that the sampling signal changes little when the electric bicycle charging load is switched on,the cumulative sum algorithm is proposed to improve the sensitivity of event detection.In order to avoid the interference of transient current pulse noise,a multifunctional composite sliding window is introduced to reduce the false detection rate of event detection.Thus,an event detection method based on composite sliding window and cumulative sum algorithm is proposed.After the event is detected,based on the principle of load superposition and separation,the voltage zero-crossing point is used as the reference point.The method separates the newly input single load steady-state current waveform,and extracts the load features to form the load sample to be identified.Based on the measured data,three scenarios are constructed under different load levels,and the results verify the effectiveness of the method.Finally,the principle of the support vector machine classifier is studied and extended to an one-class classifier.And a non-invasive electric bicycle charging load identification algorithm based on one-class support vector machine is proposed.Based on the incremental learning and the multi-core co-training method,the active learning ability of the classifier is expanded.The results show that this method effectively solves the problem that the initial classifier has a low recognition rate of unfamiliar electric bicycle load samples.Combining technical requirements and application functional requirements,a non-intrusive electric bicycle charging load identification device is designed and developed to realize functions such as data collection,event detection,feature extraction,electric bicycle charging behavior identification and alarms,which has certain practical significance.