The authors [PS7] explore the suitability of pattern classification approach of ANNs for fault detection and diagnosis. The suitability of using ANNs as pattern classifiers for power system fault diagnosis is described in detail. An ANN design and simulation environment for real-time FDD is presented. An analysis of the learning, recall and generalization characteristic of the neural network diagnostic system is presented and discussed in detail.
In the paper [PS8], two alternative approaches to the problem of fault diagnosis in real-size power systems are considered: an expert system currently in operation and a novel ANN prototype. An experimental comparison between both systems has been carried out on the basis of several test cases representing real situations. Whilst both systems perform correctly concerning functionality, the ANN prototype operates faster and is practically independent of the problem complexity.
An approach to fault diagnosis for large scale power systems is presented [PS9], based on hierarchical distributed ANNs. Several independent ANNs in the same level are used to diagnose the faults within a substation. One higher level ANN which makes use of some outputs from lower level as its inputs is used to diagnose the faults in transmission lines. A combined gradient learning algorithm with dvantages of both gradient and conjugate gradient algorithms is employed for training the ANNs. This learning algorithm converges faster than the error BP algorithm. Comparisons between the proposed approach and the single ANN approach are made for a model substation and a model power system. The authors conclude that simulation results show this approach is very encouraging.
The paper [PS10] compares the performance of two ANN models for fault diagnosis of power systems. Radial basis function and BP networks are compared with reference to generalization, training time and number of training patterns needed for each model.
The papers [PS11, PS12] propose a new connectionist (or ANN) expert diagnostic system for online fault diagnosis of a power substation. The connectionist expert diagnostic system has a similar profile to an expert system, but can be constructed much more easily from elemental samples. The User Interface here serves as a communication medium between the user and the diagnosis system, and compiles the elemental samples provided by the operator into the training samples needed. The Data Preparation Unit receives the status signals of protective relays and breakers as well as the bus voltages and feeder currents from the SCADA system. These signals are integrated in the form of symptom data and then provided to the Connectionist-type Knowledge Base and the Auxiliary Diagnosis Connectionist Model (ADCM) for diagnosis. The Connectionist-type Knowledge Base comprises there hiererchical modules of connectionists for the purposes of forward diagnosis, back tracking and comment, respectively. These three modules of connectionists can be automatically constructed according to the training sets from the User Interface. The ACDM is designed as a two-level structure of connectionists model. In the first level is a distributed processing structure. The architecture of this level consists of several connectionists, each of which is responsible for the fault diagnosis at a particular section. A section of the substation is meant in this paper by a feeder, a bus or a transformer, which can be separated by breakers. The second level of the ADCM is the Synthesis Unit. According to the output of each ADC, the Synthesis Unit provides the final diagnosis result of the ADCM. With respect to current input symptom pattern, the Inference Engine determines one of the possible fault classes presented by the connectionists in the Connectionist-type Knowledge Base and the ADCM. The decision is based on the calculation of confidence level for each possible fault class. Decision of the diagnosis can be reasonably explained by displaying the operating and failure protective devices and the associated bus voltages and feeder currents. Also, the confidence level of the answer is given to account for its credibility. The proposed approach has been practically verified by testing on a typical Taiwan power (Taipower) secondary substation. The test results suggest that this system can be implemented by various electric utilities with relatively low customization effort.
The paper [PS13] proposes a new ANN diagnostic system for online power system fault section estimation using information of relays and circuit breakers. This system has a similar profile of an expert system, but can be constructed much more easily from elemental samples. These samples associate fault section with its primary, local and/or remote protective relays and breakers. The diagnostic system can be applicable to the power system control center for single or multiple fault sections estimation, even in the cases of failure operation of relays and breakers, or error-existent data transmission. The proposed approach has been practically verified by testing on a model power system. The test results, although preliminary, suggest this system can be implemented by various electric utilities with relatively low customization effort.
The authors [PS14] present an ANN approach to a diagnostic system for a gas insulated switchgear (GIS). Firstly they survey the status of operational experience of failures in GISs and its diagnostic techniques. Secondly, they present how to acquire signal samples from the GIS and how to process them so as to be provided for an input layer of ANN. Finally they propose a decision-tree like network referred to as module ANN, and compare it with the well-known three-layered network, the straight forward.
The paper [PS15] presents an ANN approach using ART2 to a diagnostic system for GIS. To begin with, the authors show the background of abnormality diagnosis of GISs from the view point of predictive maintenance of them. Then, they discuss the necessity of ART-type ANNs, as an unsupervised learning method, in which neuron(s) are self-organized and self-created when detecting unexpected signals even if untrained by ANNs through a sensor. Finally, they present brief simulation results and their evaluation.
ANNs are used [PS16] to recognize the causes of faults in power distribution systems, based on fault currents information collected for each outage. Actual field data are used. The methodology and implementation of ANNs and fuzzy logic for the identification of animal-caused distribution faults are presented. Satisfactory results are obtained, and the developed methodology can be easily generalized and used to identify other causes of faults in power distribution systems.
The authors [PS17] describe a neural-fuzzy hybrid system to identify the causes of temporary faults as well as sustained faults. The generalization ability of the hybrid fault identification system with respect to different system configurations is analyzed and discussed in the paper.
The paper [PS18] presents a novel FFT based relaying scheme for electric utility-radial distribution high-impedance fault detection. The scheme utilizes a multi-layer feedforward ANN as a fault classifier that maps the discriminant harmonic vector of the three phase residual voltage, current and power magnitudes and phases into fault logic 1 detection. The scheme utilizes only low order harmonics, namely second, third and fifth of the residual voltage, current and power at the substation feeder transformer secondary side. The same authors present [PS20] on circuits representative of those for which a fault detection system might be applied. A method of processing time history current data has been discovered which highlights the difference between faulted and nonfault loads. An ANN has been designed and tested which utilizes the processed current data and is able with very high accuracy to distinguish faulted from nonfault loads under realistic conditions. It has been tested successfully on faults which are 1-2% of the background load on 12 kV circuits.
The paper [PS21] proposes a novel adaptive three-phase autoreclosure technique for double circuit power systems using a ANN approach. Based on the investigation of digital simulation of various types of fault on such power systems, some salient features are summarized and extracted which are then used as the inputs of ANNs. A three-layer ANN is constructed, trained and tested. The results indicate that the proposed approach is very reliable.
The paper [PS22] propose an application of ANNs in adaptive interlocking systems. Interlocks have been in use ever since protective relaying schemes were implemented for power devices like generators, transformers, transmission lines, etc. Although the science of protective relaying has undergone marked changes and improvements, the interlocking philosophy has not changed much. Recently with the availability of programmable logic controllers (PLCs), interlocking schemes have been implemented by means of these devices with basic philosophy of logic remaining the same. This paper suggests the implementation of interlocking schemes with ANNs employing threshold logic unit (TLU) elements. It is demonstrated that while the basic hardware required is same as that of any common PLC, the suggested system will have added flexibility, adaptability to various switchyard modifications, electrical topology changes and equipment/switchyard conditions as well as network complexity.
In the paper [PS23] a typical problem of maloperation is considered. The application of the modified multilayer perceptron (MLP) mode can successfully avoid the maloperation of a relay. For the cases considered, it shows encouraging results. The advantage associated with the presented MLP model is that the modified characteristic can be defined in the absence of a definite analytical model since the ANN can learn it through input-output patterns. The methodology can be extended to many adaptive protective schemes. This report just opens new vistas for the exploration of the application of ANNs in adaptive protective schemes, and further investigations could lead to increased confidence.
The paper [PS24] reports about the application of ANNs as nonlinear filters. The ANNs are used to restore current waveforms distorted by saturation of current transducers. The paper presents the progress in this application of ANN.
In the paper [PS25] the performance of an ANN with an activation function with constant and adaptable slope is presented. Later the network can be used as a neurocontroller to estimate the state and isolate the faulty power system from the rest of the system. The method is illustrated with an example.
The authors [PS26] propose an adaptive protection technique for controllable series compensated EHV lines. As is well known, flexible AC transmission systems (FACTS) provide opportunities of better utilizing existing transmission systems by using power electronics based controllers. One of the main FACTS devices is controllable series compensation (CSC), which has an ability to control the compensated impedance by changing the firing angle of thyristors. However, the implementation of this technology will pose new problems to conventional line protection schemes. This paper proposes a novel adaptive protection scheme for CSC transmission systems by using an ANN approach. It places emphasis on the feature extraction, the topology and training of ANNs. Some preliminary test results clearly show the trained ANN is able to make correct trip decisions from abnormal voltage waveforms using associations learned from previous experiences. In addition, the scheme also has the ability to identify faulted phases. The test results successfully demonstrate the feasibility of ANNs based adaptive protection for CSC transmission systems.
A series compensation technique has been employed [PS27] to improve power transfer in long-distance transmission systems world-wide. However, this in turn introduces problems in conventional distance protection. The complex variation of line impedance is accentuated as the capacitor's own protection equipment operates randomly under fault conditions. This paper proposes a novel adaptive protection scheme for a series compensated transmission system by using an ANN approach.
Send Questions, Comments, and Suggestions to bencr@hotmail.com