Study On Intellegent Fault Diagnosis And Harmonic Source Identification In Power Distribution Network

Fault diagnosis is the foundation of precise fault analysis and fast fault restoration. Harmonic source identification is the premise of harmonic treatment and responsibility partition of harmonic pollution. With the development of science and technology and the progress of society, the size of distribution network keeps growing, and protection and monitor devices are more installed. As a result, large amount of data was accepted by information center in nowadays. How to use these data for fault diagnosis and harmonic source identification in distribution network is both theoretically and practically significant for improving operation level in power system. According to the characteristics of fault diagnosis and harmonic source identification in distribution, the effective electrical components are selected, and the reasonable techniques are employed for studying the related issues.At first, the fault classification technique in neutral non-effectively grounded distribution network based on Adaptive Network-based Fuzzy Inference System (ANFIS) is proposed. The transient signals in fault characteristic frequency band are extracted by wavelet transform. From the transient signals, the fault characteristic quantities are constructed by statistics. Afterward, their performance under different fault types is studied. ANFIS is employed as the fault characteristic quantities fusion tool to obtain the final fault classification result. The proposed fault classification technique is trained and tested in the simulated distribution network in PSCAD/EMTDC environment. The results show its high correctness. The adaptability of the proposed method is studied in six distribution operating conditions, which are neutral grounded style changing, arc fault, different system equivalent impedances, network topology variation, noise interference and different load levels. From the results, the classification method is almost immune in the first four conditions. When nework topology changes slightly, the adaptability is good. But, the correctness of results would decrease if topology changes intensively. The proposed technique is seriously degraded by small signal noise ratio or heavy load in the system. So, adding noise filter and training samples of heavy load should be considered for improving the correctness. At second, the fault line selection methods based on S-Transform (ST) in neutral non-effectively grounded distribution network is well studied. Through analyzing the amplitude-frequency and phase-frequency characteristics of ST, the fault line selection technique based on phase-comparison principle is proposed. This technique utilizes the modulus and phase information extracted by ST at different frequencies of zero sequence currents. Through modulus comparison, the characteristic frequency is found. Through phase comparison, the fault line vote mechanism is constructed. Since the modulus at characteristic frequency can reflect the reliability of phase at the corresponding frequency, the vote confidence degree is defined in vote process. Through one by one vote in1/4fundamental cycle after fault, the vote statistic diagram is obtained. The Fault Line Selecting Confidence Degree (FLSCD), which is defined by Shannon Fuzzy Entropy (SFE), is used to calculate the vote statistic diagram. In the proposed fault line selection method, the fault line result is given with FLSCD. Its advantages are that the result reliability is highly enhanced by utilizing multiple-points vote results, and FLSCD is the reliability measument of the suspected fault line. The simulation model with four feeders is established. The simulating results show that the proposed technique possesses high correctness, and the defined FLSCD is rational. Later, the difference between instant power of faulty line and that of healthy lines are studied. The instant power functions for fault line selection based on ST are constructed for proposing the other fault line selection method. The performance of two above methods in arc fault and noise environment is researched. The results show that both of them exibihit good in arc fault, and their anti-noise ability is also strong. In the end, two methods are synthetized for obtaining the fusion method based on SFE. Through simulation, the anti-noise ability of fusion method is improved in comparison with the single principle based methods.At third, the main factors which attenuate the fault location signal in S-injection method are theoretically studied. Moreover, these factors are researched by simulation. It is found that fault resistance has the most serious influence on the fault location signal. When the network topology is fixed, the attenuation of fault location signal mainly depends on fault resistance and fault distance. To make the railway distribution network as application object, the fault location system based on S-injection method is developed. It employs wireless detectors which are based on the communication idea of wireless sensor network to upload the sensing information of fault location signal. As a result, the efficiency of tracing location signal is highly improved. From the field test on a real feeder, the location system locates fault correctly and quickly.At last, the harmonic source identification in power distribution is researched. The harmonic power flow direct algorithm based on current injection idea is modified first. The algorithm can calculate the network with loops, capacitors and filters through matrix initial transform. In comparison with the results of distribution in IEEE-519, the correctness of algorithm is verified. Compared with backward-forward sweep method, the algorithm in this paper shows fast calculation speed and good convergence. Moreover, its advantages are more obvious as the number of loops increases. Then, two harmonic source identification methods based on Least-Square Estimation (LSE) and Sparsity Maximizatioin (SM) respectively are studied carefully. According to IEEE-123test network, the calculation model for testing the two methods is established. The identification ability of two methods is analyzed when there are different numbers of measument devices in network. The adaptability of two techniques is compared when non-main harmonic sources have injecting disturbances, measument matrix has errors, and network has loops or capacitors. From the study results, the LSE method shows better adaptability, however the expense is more measument devices. The SM technique can identify harmonic sources with the small number of measument devices. Also, it has good adaptability under the conditions of harmonic injection distrubance, loops and inaccurate measument matrix. But when capacitors are connected into network, the SM technique is degraded seriously or even invalid. So, the harmonic current flowing in capacitors should be measured in order to avoid their impact.