Fault Forecast And Dignosis For Low-Voltage Switchgear Based On Fuzzy Theory And RBF Neural Network

The safety of the low-voltage switchgear affects the running of the whole of the power supply system directly. In order to avoid planned maintenance bringing to overmuch maintenance and deficiency maintenance, this paper does nether measures: Predict working state measures of low-voltage switching apparatus by fuzzy algorithm. According to health state of device, make up maintain plan. Then establish intelligent fault diagnosis model-RBF neural network to diagnose the faults of device. In this way, save maintenance cost, but also reduce the fault checking time.Firstly, the structure of the low-voltage switchgear cabinet is analysised. Take the water pump drawer cell for example. Because of faults occurred on apparatus, analyze parameters of prediction and faults. In view of importance of apparatus and failure rate, this paper mainly studies circuit breaker.Subsequently, study fuzzy comprehensive evaluation. Point out disadvantages about common methods of weight. Integrate trigonometry fuzzy into AHP method to determine weight coefficient of fault forecast. To some extent, adjust deviation of construction process about judgment matrix and its process accord with people's thinking and expression. This improved fuzzy comprehensive evaluation is applied to fault prediction of circuit breaker. The simulation results show the method be efficient.Thirdly, analyze off-line training and on-line training of RBF neural network.this paper puts forward hybrid training method based on nearest neighbor-clustering algorithm and gradient descent algorithm. This RBF neural network based on hybrid training method(nearest neighbor-clustering algorithm and gradient descent algorithm) is applied to incidental mechanical faults of ciecuit breaker.A simulation experiment in MATLAB is carried out. Accoriding to comparison about RBF neural network based on the method nearest neighbor-clustering algorithm and RBF neural network based on gradient descent algorithm, the diagnosis results of RBF neural network based on based on nearest neighbor-clustering algorithm and gradient descent algorithm is better and the diagnosis is very precise.