Study On The Progressive Changes Analysis Method Of Transformation Equipments State Based On Electrical Information

The condition of transformation equipments plays an important role on the safe and reliable operation of the power grid, and its effect becomes more obvious with the development of power grid and the electrical equipments capacity. Once the fault of transformation equipments occurs, it will bring customer outage and economic losses, even threat to personal safety. Therefore, study on the potential fault detection, the condition evaluation and maintenance measures of transformation equipments will be of great importance.Traditional preventive maintenance system is gradually replaced by the condition-based maintenance system, due to its high costs and poor ability to detect latent failures. Condition-based maintenance is based on online monitoring and evaluation of the equipments condition, to determine the appropriate maintenance time and maintenance measures and achieve the optimal allocation of human, financial and material resources. The main ideas of the present researches concentrate on comprehensive analysis of electrical and non-electrical data and using a variety of mathematical algorithms to assess equipments condition. However, condition-based maintenance pays more attention to the changing of equipment state which is the basis to determine the accurate maintenance time, maximize equipment utilization and minimize the impact on normal production. The existed studies have focused on the current health assessment of equipments, and lack the analysis of the gradual changing process of equipments state, so it is imperative to study the appropriate analysis methods. The method that provides assisted analysis information for equipments condition assessment by using abundant electrical information has advantages in no additional installed device, rich measurement data and convenient access. Therefore, based on electrical information and from the perspective of data mining, this dissertation compares the probability distribution of equipments port model parameters to extract gradual process features and analyze the impact on transformation equipments state changing process from lightning and external short circuit failure. And based on this, the characteristic of future change tendency is extracted, so more information to assist the development of maintenance measures and to be conducive to the further implementation of condition-based maintenance is provided.The innovative work of this dissertation are as follows: (1) The analysis method of transformation equipments state progressive changes: the process of transformation equipments state progressive changes is the accumulation of many small changes, and these small changes can be reflected indirectly by the parameters variation of equipments port model built by generalized volt-ampere characteristics. However, due to the impact of the external environment and measurement errors, the corresponding parameter identification results show strong randomness and its inherent characteristics are difficult to be extracted. Therefore, the progressive changes analysis method based on the statistical theory is proposed. First, the equipments operation process is divided into many intervals and the distribution of port model parameters in each interval is analyzed based on its probability density function calculated by nonparametric kernel estimation method. Then, the probability distribution differences of parameters in different internals are got and four indexes presenting the changing process are defined. These indexes are: the parameter value Ckmax whose probability is maximum; the difference between Ckmax and C1max; the parameter change probability relative to the first interval; the parameter change probability relative to the warning status. Finally, the changing process of port model parameters is analyzed using these indexes, and the sequences can be got, which can provide auxiliary analysis for equipments state trend. The proposed method based on statistics, analyzes the progressive changes through lots of historical samples, which is affected little by bad data and has strong anti-distrubance ability and great robustness. In the analysis process, the probability density function can be calculated by nonparametric kernel estimation method, which does not presuppose parameter distribution, and the parameters are identified by partial least squares regression algorithm. Taking the analysis of the cumulative effects of transformer winding deformation as an example, the leakage inductance is got by Monte Carlo simulation and the port model parameter changing process is analyzed by defined indexes. The analysis results show that this analysis method is effective and feasible.(2) The impact analysis of shocks to the equipments state progressive changes: In the operation of transformation equipments, the lightning strikes, short-circuit faults and other shocks, may cause equipments state change, which will threat to the safety of equipments. And the quantitative analysis of port model parameter changes is necessary to study its further changing process. However, the random fluctuation caused by environment and measurement errors increase the detection difficulties. Therefore, considering the fluctuation characteristics, two analysis methods are put forward, which based on probability density differences and adaptive integral algorithm respectively. In the former method, the parameter change is detected by the probability function difference in adjacent time windows and the variation amplitude is represented by the difference value of the parameters whose probability is maximum. This method is accurate and requires large amount of calculation, which is suitable to analyze the state changes after shocks. In the latter method, the parameter changes are detected by analyzing the change of the mean value of the difference samples, and the variation is represented by the mean value. The method has quick calculation ability and can detect the mutation timely that does not reach the action condition of protective devices. In the two methods, the threshold value is determined adaptively, which can realize the coordination of the detection sensitivity and accuracy. Change the leakage inductance to simulate the transformer state changes caused by lighting, short-circuit faults and other shocks, the analysis results show that the two methods are effective and reliable.(3) The trend analysis of port model parameter changes which reflect equipments state indirectly:In condition-based maintenance, the equipments state changing trend should be analyzed in long time scale and short time scale. For this purpose, according to the index sequences representing the parameter changing process, the trend component is extracted by Empirical Mode Decomposition, the index prediction model in long time scale is built, and the interval that the index reaches the value which is corresponding to warning status is estimated, which can provide auxiliary analysis for the evaluation of the time distance between the current state and the warning status. This is helpful to assess equipments condition and make scientific maintenance measures. To analyze the port model parameter changes in detail, a method based on state transition probability matrix is presented. By analyzing the distribution of parameter samples in two adjacent time windows, the probability transition probability matrix is established and the parameter in future time windows can be predicted, which can provide the auxiliary analysis for the equipments state changing details. Taking the analysis of the transformer state changes caused by the cumulative effect of transformer winding as an example, based on the indexes representing port model parameter changes, the time that the index reaches the value corresponding to warning state is evaluated, which can provide auxiliary information for transformer condition assessment and making maintenance measures. To simulate the transformer close to warning state, the leakage inductance samples are derived by Monte Carlo simulation method. The state transition probability matrix is calculated by the samples in the first two windows, and parameter distribution in the third window is predicted. The similarity between the predicted and simulated samples is high, which shows that the method is effective.