技术领域technical field
本发明涉及电力系统输变电设备运行维护领域中的变压器状态参量预测方法及系统，尤其涉及一种基于果蝇算法优化的变压器状态参量预测方法及系统。The invention relates to a transformer state parameter prediction method and system in the field of operation and maintenance of power transmission and transformation equipment in a power system, in particular to a transformer state parameter prediction method and system optimized based on a fruit fly algorithm.
背景技术Background technique
电力变压器作为输变电系统的主要核心设备，保证其能够健康稳定地运行具有重要意义。通常通过监测变压器运行状态并预测其变化趋势，从而有效预防变压器故障、做好预案，保证变压器的稳定运行。为了有效监测变压器运行状态并预测其变化趋势，通常需要监测反映变压器运行状态的各变压器状态参量并对其进行预测。As the main core equipment of power transmission and transformation system, power transformer is of great significance to ensure its healthy and stable operation. Usually, by monitoring the operating status of the transformer and predicting its changing trend, it can effectively prevent transformer failures and make plans to ensure the stable operation of the transformer. In order to effectively monitor the operating state of the transformer and predict its changing trend, it is usually necessary to monitor and predict the state parameters of the transformer that reflect the operating state of the transformer.
变压器的状态参量信息维度很高，数据量庞大，不同的状态参量数据从不同角度在一定程度上反映了变压器的运行状态。变压器的状态参量数据主要包括在线监测数据、离线实验数据、运行巡检数据以及变压器自身技术参数数据等，这些数据可以从各个方面反映变压器的运行状态，利用这些数据对变压器运行状态进行预测具有重要的研究意义。变压器在线监测状态参量数据是一个复杂的数据序列，对变压器在线监测状态参量数据未来的变化趋势进行预测，能够更好地预测变压器的运行状态变化。The state parameter information of the transformer has a high dimension and a huge amount of data. Different state parameter data reflects the operating state of the transformer to a certain extent from different angles. The state parameter data of the transformer mainly includes online monitoring data, offline experimental data, operation inspection data and the transformer's own technical parameter data, etc. These data can reflect the operating state of the transformer from various aspects. research significance. Transformer online monitoring state parameter data is a complex data sequence. Predicting the future change trend of transformer online monitoring state parameter data can better predict the change of transformer operating state.
传统的变压器状态参量预测模型仅仅考虑了单一变量或者少数几个变量的数据对序列未来的变化趋势进行判断。并且在对历史数据进行拟合时，无法保留较为久远的历史信息对当前时刻的作用，从而无法实现对未来较长时间尺度的预测。The traditional transformer state parameter prediction model only considers the data of a single variable or a few variables to judge the future change trend of the sequence. Moreover, when fitting historical data, it is impossible to retain the effect of relatively long-term historical information on the current moment, so that it is impossible to predict a long time scale in the future.
广义回归神经网络(generalized regression neural network,GRNN) 是由美国学者Donld F.Specht提出的一种新型的径向基神经网络。它是以非线性回归分析为基础的径向基神经网络。该网络有能力从历史数据中预测任意形式的函数，并且最后收敛于样本量积累最多的优化回归面。在结构上和多层感知神经网络类似，包括输入层、模式层、求和层和输出层。广义回归神经网络在网络的训练过程中不需要调整每个神经元之间的连接的权重，网络的学习过程仅仅依赖于数据样本本身，模型唯一需要设定的是光滑因子参数，很大程度上降低了人为主观影响因素，因此特别适用于具有非线性特征的变压器状态参量数据的预测。Generalized regression neural network (GRNN) is a new type of radial basis neural network proposed by American scholar Donld F. Specht. It is a radial basis neural network based on nonlinear regression analysis. The network has the ability to predict functions of any form from historical data, and finally converges to the optimal regression surface with the largest accumulation of samples. Similar in structure to a multi-layer perceptual neural network, it includes an input layer, a pattern layer, a summation layer, and an output layer. The generalized regression neural network does not need to adjust the weight of the connection between each neuron during the training process of the network. The learning process of the network only depends on the data sample itself. The only thing that the model needs to set is the smooth factor parameter. It reduces human subjective influence factors, so it is especially suitable for the prediction of transformer state parameter data with nonlinear characteristics.
基于神经网络构建的预测模型可以用于变压器状态参量数据预测，但是存在超参数选取容易陷入局部收敛，导致预测模型训练效率低，影响预测准确率和可靠性的问题。The prediction model based on neural network can be used for the prediction of transformer state parameter data, but there is a problem that the selection of hyperparameters is easy to fall into local convergence, resulting in low training efficiency of the prediction model and affecting the prediction accuracy and reliability.
发明内容SUMMARY OF THE INVENTION
本发明的目的之一是提供一种基于果蝇算法优化的变压器状态参量数据预测方法，该方法能避免超参数选取陷入局部收敛，从而提升预测模型训练效率，保证较高的变压器状态参量数据预测准确率和可靠性。One of the objectives of the present invention is to provide a method for predicting transformer state parameter data based on fruit fly algorithm optimization, which can avoid the local convergence of hyperparameter selection, thereby improving the training efficiency of the prediction model and ensuring higher transformer state parameter data prediction. Accuracy and reliability.
根据上述发明目的，本发明提出了一种基于果蝇算法优化的变压器状态参量数据预测方法，其对变压器状态参量数据进行预测，所述方法包括以下步骤：According to the above purpose of the invention, the present invention proposes a method for predicting transformer state parameter data based on fruit fly algorithm optimization, which predicts transformer state parameter data, and the method includes the following steps:
S100：获取一段时间内的变压器状态量数据，并将其转换为矩阵形式的变压器状态量矩阵，所述变压器状态量包括变压器状态参量的相关数据；S100: Acquire transformer state quantity data within a period of time, and convert it into a transformer state quantity matrix in the form of a matrix, where the transformer state quantity includes relevant data of transformer state parameters;
S200：构建变压器状态参量数据预测模型，基于果蝇算法求得所述预测模型的超参数，基于所述变压器状态量矩阵对所述预测模型进行训练；S200: construct a transformer state parameter data prediction model, obtain hyperparameters of the prediction model based on the Drosophila algorithm, and train the prediction model based on the transformer state parameter matrix;
S300：基于经步骤S200训练的变压器状态参量数据预测模型预测变压器状态参量数据。S300: Predict the transformer state parameter data based on the transformer state parameter data prediction model trained in step S200.
本发明提出的基于果蝇算法优化的变压器状态参量数据预测方法，其采用果蝇算法求得预测模型的超参数，以避免超参数选取陷入局部收敛，从而提升预测模型训练效率，保证较高的变压器状态参量数据预测准确率和可靠性。超参数是训练之前设置值的参数，而不是通过训练得到的参数值。通常情况下，需要对超参数进行优化，选择一组最优超参数，以提高训练的效率和效果。果蝇算法(Fruit Fly Optimization Algorithm，FOA)是由中国台湾学者Wen-Tsao Pan提出的一种优化算法，该算法的思路是模拟果蝇觅食的行为：首先以嗅觉发现事物的大致方位，然后再在近距离的情况下用视觉发现事物的具体位置并接近觅食的过程。果蝇算法具有很好的全局优化性能，因此基于果蝇算法求得的预测模型的超参数，能避免超参数选取陷入局部收敛，从而提升预测模型训练效率，保证较高的变压器状态参量数据预测准确率和可靠性。The transformer state parameter data prediction method based on the fruit fly algorithm optimization proposed by the present invention adopts the fruit fly algorithm to obtain the hyperparameters of the prediction model, so as to avoid local convergence in the selection of hyperparameters, thereby improving the training efficiency of the prediction model and ensuring a higher Transformer state parameter data prediction accuracy and reliability. A hyperparameter is a parameter whose value is set before training, not a parameter value obtained through training. Usually, hyperparameters need to be optimized, and a set of optimal hyperparameters is selected to improve the efficiency and effect of training. The Fruit Fly Optimization Algorithm (FOA) is an optimization algorithm proposed by Taiwanese scholar Wen-Tsao Pan. The idea of the algorithm is to simulate the foraging behavior of fruit flies: first, the general orientation of things is found by smell, and then Then in the case of close range, use vision to find the specific location of things and approach the process of foraging. The Drosophila algorithm has good global optimization performance. Therefore, the hyperparameters of the prediction model obtained based on the Drosophila algorithm can avoid local convergence in the selection of hyperparameters, thereby improving the training efficiency of the prediction model and ensuring higher data prediction of transformer state parameters. Accuracy and reliability.
通常所述预测模型基于神经网络构建。Typically the predictive model is built on the basis of neural networks.
进一步地，本发明所述的基于果蝇算法优化的变压器状态参量数据预测方法中，步骤S200中所述变压器状态参量数据预测模型基于广义回归神经网络构建，所述超参数为所述广义回归神经网络中的光滑因子。Further, in the method for predicting transformer state parameter data based on fruit fly algorithm optimization of the present invention, the transformer state parameter data prediction model in step S200 is constructed based on a generalized regression neural network, and the hyperparameter is the generalized regression neural network. Smoothing factor in the network.
上述方案中，广义回归神经网络是以非线性回归分析为基础的径向基神经网络。该网络有能力从历史数据中预测任意形式的函数，并且最后收敛于样本量积累最多的优化回归面；广义回归神经网络在网络的训练过程中不需要调整每个神经元之间的连接的权重，网络的学习过程仅仅依赖于数据样本本身，模型唯一需要设定的是光滑因子参数，很大程度上降低了人为主观影响因素，因此特别适用于具有非线性特征的变压器状态参量数据的预测。此外，广义回归神经网络可以保留较为久远的历史信息对当前时刻的作用，从而可以实现对未来较长时间尺度的预测。所述光滑因子为所述广义回归神经网络所要设置的唯一超参数。In the above scheme, the generalized regression neural network is a radial basis neural network based on nonlinear regression analysis. The network has the ability to predict functions of any form from historical data, and finally converges to the optimal regression surface with the most accumulated samples; the generalized regression neural network does not need to adjust the weight of the connection between each neuron during the network training process , the learning process of the network only depends on the data sample itself, and the only thing that the model needs to set is the smooth factor parameter, which greatly reduces the human subjective influence factor, so it is especially suitable for the prediction of transformer state parameter data with nonlinear characteristics. In addition, the generalized regression neural network can retain the effect of the longer historical information on the current moment, so that the prediction of the longer time scale in the future can be realized. The smoothing factor is the only hyperparameter to be set by the generalized regression neural network.
进一步地，本发明所述的基于果蝇算法优化的变压器状态参量数据预测方法中，步骤S200中所述果蝇算法采用动态步长初始化群体中个体的位置。Further, in the method for predicting transformer state parameter data based on Drosophila algorithm optimization according to the present invention, the Drosophila algorithm in step S200 uses dynamic steps to initialize the positions of individuals in the group.
传统方果蝇算法采用固定步长初始化群体中个体的位置。固定步长存在以下问题：如果设定的步长过大，则会导致算法的搜索能力变弱，花费的搜索时间过长，导致算法的效率偏低；如果设置的步长过小，则算法容易陷入局部最优。因此上述方案考虑以动态步长代替传统的固定步长，从而避免上述固定步长存在的问题。The traditional Drosophila algorithm uses a fixed step size to initialize the positions of individuals in the population. The fixed step size has the following problems: if the set step size is too large, the search ability of the algorithm will be weakened, and the search time will be too long, resulting in low efficiency of the algorithm; if the set step size is too small, the algorithm will It is easy to fall into local optimum. Therefore, the above scheme considers replacing the traditional fixed step size with a dynamic step size, so as to avoid the above-mentioned problems of the fixed step size.
进一步地，本发明所述的基于果蝇算法优化的变压器状态参量数据预测方法中，步骤S200中所述果蝇算法采用交叉验证法，将种群分为多个相等的子种群，然后分别进行优化分析，最终再选择最优解。Further, in the method for predicting transformer state parameter data based on the fruit fly algorithm optimization of the present invention, the fruit fly algorithm in step S200 adopts the cross-validation method, divides the population into multiple equal sub-populations, and then optimizes them respectively. Analysis, and finally choose the optimal solution.
上述方案保证了算法能够充分利用数据，防止算法陷入局部最优解。The above scheme ensures that the algorithm can make full use of the data and prevents the algorithm from falling into a local optimal solution.
进一步地，本发明所述的基于果蝇算法优化的变压器状态参量数据预测方法中，步骤S200中采用误差反向传播算法对所述预测模型进行训练。Further, in the method for predicting transformer state parameter data based on fruit fly algorithm optimization of the present invention, in step S200, an error back propagation algorithm is used to train the prediction model.
上述方案中，所述预测模型的结构参数通常包括广义回归网络的隐藏层层数、神经元节点数，所述误差反向传播算法可以用于确定所述预测模型的结构参数。In the above solution, the structural parameters of the prediction model generally include the number of hidden layers and the number of neuron nodes of the generalized regression network, and the error back-propagation algorithm can be used to determine the structural parameters of the prediction model.
进一步地，本发明所述的基于果蝇算法优化的变压器状态参量数据预测方法中，步骤S100中所述变压器状态参量的相关数据包括参量本身的数据和 /或参量间的比值数据。Further, in the method for predicting transformer state parameter data based on fruit fly algorithm optimization of the present invention, the relevant data of the transformer state parameter in step S100 includes the data of the parameter itself and/or the ratio data between the parameters.
上述方案中，为了进一步保证预测效果，通常希望多纳入一些维度的变压器状态量，即状态量选择范围不限于状态参量，还可以包括状态参量之间的比值等相关数据。In the above solution, in order to further ensure the prediction effect, it is usually desirable to include more transformer state quantities in some dimensions, that is, the selection range of state quantities is not limited to state parameters, and can also include related data such as ratios between state parameters.
进一步地，本发明所述的基于果蝇算法优化的变压器状态参量数据预测方法中，步骤S100中所述变压器状态参量包括变压器本体参量和变电站环境参量。Further, in the method for predicting transformer state parameter data based on fruit fly algorithm optimization of the present invention, the transformer state parameters in step S100 include transformer body parameters and substation environment parameters.
上述方案中，由于变压器运行状态不仅受变压器本体参量的影响，还受变电站环境参量的影响，因此将变压器本体参量和变电站环境参量纳入变压器状态参量的选择范围。In the above scheme, since the operating state of the transformer is not only affected by the parameters of the transformer itself, but also by the environmental parameters of the substation, the parameters of the transformer itself and the environmental parameters of the substation are included in the selection range of the transformer state parameters.
更进一步地，上述基于果蝇算法优化的变压器状态参量数据预测方法中，所述变压器本体参量包括变压器油中溶解的气体含量和/或变压器油温。Further, in the above-mentioned method for predicting transformer state parameter data based on Drosophila algorithm optimization, the parameters of the transformer body include the dissolved gas content in the transformer oil and/or the transformer oil temperature.
上述方案中，变压器油中溶解的气体含量和变压器油温在一定程度上可以反映变压器绝缘老化或故障的程度，因此将变压器油中溶解的气体含量和变压器油温纳入变压器本体参量的选择范围。In the above scheme, the dissolved gas content in the transformer oil and the transformer oil temperature can reflect the degree of transformer insulation aging or failure to a certain extent. Therefore, the dissolved gas content in the transformer oil and the transformer oil temperature are included in the selection range of the transformer body parameters.
更进一步地，上述基于果蝇算法优化的变压器状态参量数据预测方法中，所述变电站环境参量包括气温、地面湿度、相对湿度、平均风速中的一种或多种。Further, in the above-mentioned method for predicting transformer state parameter data based on fruit fly algorithm optimization, the substation environmental parameters include one or more of air temperature, ground humidity, relative humidity, and average wind speed.
上述方案中，气温、地面湿度、相对湿度、平均风速等因素会对变压器的性能造成影响，被认为是变压器状态的相关因素，因此将气温、地面湿度、相对湿度、平均风速等因素纳入变压器本体参量的选择范围。In the above scheme, factors such as temperature, ground humidity, relative humidity, and average wind speed will affect the performance of the transformer and are considered to be related factors of the transformer state. Therefore, factors such as temperature, ground humidity, relative humidity, and average wind speed are included in the transformer body. Parameter selection range.
更进一步地，上述基于果蝇算法优化的变压器状态参量数据预测方法中，所述变压器油中溶解的气体包括H2、CO、CH4、C2H2、C2H4、C2H6中的一种或多种。Further, in the above-mentioned method for predicting transformer state parameter data based on Drosophila algorithm optimization, the gas dissolved in the transformer oil includes H2 , CO, CH4 , C2 H2 , C2 H4 , C2 H6 one or more of.
上述方案中，可以基于油中溶解气体分析(DGA)相应的色谱数据得到相关参量的原始数据。上述方案采选择的气体种类包括氢气(H2)、一氧化碳(CO)、甲烷(CH4)、乙炔(C2H2)、乙烯(C2H4)以及乙烷(C2H6)。In the above solution, the raw data of the relevant parameters can be obtained based on the corresponding chromatographic data of the Dissolved Gas Analysis (DGA) in the oil. The gas species selected for the above scheme include hydrogen (H2 ), carbon monoxide (CO), methane (CH4 ), acetylene (C2 H2 ), ethylene (C2 H4 ) and ethane (C2 H6 ).
更进一步地，上述基于果蝇算法优化的变压器状态参量数据预测方法中，基于所述变压器油中溶解的气体，所述参量间的比值数据包括CH4/H2、 C2H2/C2H4、C2H4/C2H6、C2H6/CH4、C2H2/CH4、C2H6/C2H2中的一种或多种。Further, in the above-mentioned method for predicting transformer state parameter data based on Drosophila algorithm optimization, based on the gas dissolved in the transformer oil, the ratio data between the parameters includes CH4 /H2 , C2 H2 /C2 One or more of H4 , C2 H4 /C2 H6 , C2 H6 /CH4 , C2 H2 /CH4 , C2 H6 /C2 H2 .
上述方案中，为了拓展输入向量的维度，借鉴了利用油中溶解气体分析 (DGA)比值对变压器故障进行诊断的思路，将常用的国际电工委员会(IEC) 比值、Rogers比值以及Dornenburg比值也纳入输入向量的选择范围，即将 CH4/H2、C2H2/C2H4、C2H4/C2H6、C2H6/CH4、C2H2/CH4、C2H6/C2H26组DGA比值也纳入所述参量间的比值数据选择范围。In the above scheme, in order to expand the dimension of the input vector, the idea of using the dissolved gas analysis (DGA) ratio to diagnose transformer faults is used for reference, and the commonly used International Electrotechnical Commission (IEC) ratio, Rogers ratio and Dornenburg ratio are also included in the input. The selection range of the vector, namely CH4 /H2 , C2 H2 /C2 H4 , C2 H4 /C2 H6 , C2 H6 /CH4 , C2 H2 /CH4 , C2 The H6 /C2 H2 6 group DGA ratio is also included in the data selection range of the ratio between the parameters.
本发明的另一目的是提供一种基于果蝇算法优化的变压器状态参量数据预测系统，该系统能能避免超参数选取陷入局部收敛，从而提升预测模型训练效率，保证较高的变压器状态参量数据预测准确率和可靠性。Another object of the present invention is to provide a transformer state parameter data prediction system optimized based on Drosophila algorithm, which can avoid the local convergence of hyperparameter selection, thereby improving the training efficiency of the prediction model and ensuring higher transformer state parameter data. Forecast accuracy and reliability.
根据上述发明目的，本发明提出了一种基于果蝇算法优化的变压器状态参量数据预测系统，该系统包括数据连接的数据采集模块和数据处理模块，采用上述变压器状态参量数据预测方法中的任意一种方法预测变压器状态参量数据。According to the above purpose of the invention, the present invention proposes a transformer state parameter data prediction system optimized based on the fruit fly algorithm. The system includes a data acquisition module and a data processing module connected with data, and adopts any one of the above transformer state parameter data prediction methods. A method to predict transformer state parameter data.
本发明提出的基于果蝇算法优化的变压器状态参量数据预测系统，其通过采用上述任一变压器状态参量数据预测方法对变压器状态参量数据进行预测，因此，根据前述原理，该系统能避免超参数选取陷入局部收敛，从而提升预测模型训练效率，保证较高的变压器状态参量数据预测准确率和可靠性。The transformer state parameter data prediction system based on the fruit fly algorithm optimization proposed by the present invention predicts the transformer state parameter data by adopting any of the above transformer state parameter data prediction methods. Therefore, according to the aforementioned principle, the system can avoid hyperparameter selection. It falls into local convergence, thereby improving the training efficiency of the prediction model and ensuring a higher prediction accuracy and reliability of the transformer state parameter data.
本发明所述的基于果蝇算法优化的变压器状态参量数据预测方法能能避免超参数选取陷入局部收敛，从而提升预测模型训练效率，保证较高的变压器状态参量数据预测准确率和可靠性。本发明方法采用果蝇算法优化和预测模型的组合，具有很好的灵活性和拓展性，可以自由选择合适的预测模型。本发明方法经实验验证具有较好的拟合能力和预测能力。The method for predicting transformer state parameter data based on fruit fly algorithm optimization can avoid the local convergence of hyperparameter selection, thereby improving the training efficiency of the prediction model and ensuring higher prediction accuracy and reliability of transformer state parameter data. The method of the invention adopts the combination of fruit fly algorithm optimization and prediction model, has good flexibility and expansibility, and can freely select a suitable prediction model. The method of the invention has better fitting ability and prediction ability after experimental verification.
本发明所述的基于果蝇算法优化的变压器状态参量数据预测系统同样具有以上优点和有益效果。The transformer state parameter data prediction system optimized based on the fruit fly algorithm of the present invention also has the above advantages and beneficial effects.
附图说明Description of drawings
图1为本发明所述的基于果蝇算法优化的变压器状态参量数据预测方法的流程示意图。FIG. 1 is a schematic flowchart of the method for predicting transformer state parameter data based on fruit fly algorithm optimization according to the present invention.
图2为本发明所述的基于果蝇算法优化的变压器状态参量数据预测方法在一种实施方式下的流程示意图。FIG. 2 is a schematic flowchart of the method for predicting transformer state parameter data based on fruit fly algorithm optimization according to an embodiment of the present invention.
图3为验证实例中光滑因子优化过程的均方根误差Rmse变化曲线示意图。FIG. 3 is a schematic diagram of the variation curve of the root mean square error Rmse of the smoothing factor optimization process in the verification example.
图4为验证实例中甲烷气体浓度预测结果相对百分误差示意图。FIG. 4 is a schematic diagram of the relative percentage error of the prediction result of the methane gas concentration in the verification example.
具体实施方式Detailed ways
下面将结合说明书附图和具体的实施例对本发明所述的基于果蝇算法优化的变压器状态参量数据预测方法及系统做进一步的详细说明。The method and system for predicting transformer state parameter data based on fruit fly algorithm optimization according to the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.
图1示意了基于果蝇算法优化的变压器状态参量数据预测方法的流程。Figure 1 illustrates the flow of the transformer state parameter data prediction method optimized based on the Drosophila algorithm.
如图1所示，本发明的基于果蝇算法优化的变压器状态参量数据预测方法的流程包括：As shown in Figure 1, the flow of the transformer state parameter data prediction method optimized based on the fruit fly algorithm of the present invention includes:
S100：获取一段时间内的变压器状态量数据，并将其转换为矩阵形式的变压器状态量矩阵，所述变压器状态量包括变压器状态参量的相关数据；S100: Acquire transformer state quantity data within a period of time, and convert it into a transformer state quantity matrix in the form of a matrix, where the transformer state quantity includes relevant data of transformer state parameters;
S200：构建变压器状态参量数据预测模型，基于果蝇算法求得所述预测模型的超参数，基于所述变压器状态量矩阵对所述预测模型进行训练；S200: construct a transformer state parameter data prediction model, obtain hyperparameters of the prediction model based on the Drosophila algorithm, and train the prediction model based on the transformer state parameter matrix;
S300：基于经步骤S200训练的变压器状态参量数据预测模型预测变压器状态参量数据。S300: Predict the transformer state parameter data based on the transformer state parameter data prediction model trained in step S200.
在某些实施方式中，步骤S300中变压器状态参量数据预测模型基于广义回归神经网络构建，所述超参数为所述广义回归神经网络中的光滑因子。In some embodiments, the transformer state parameter data prediction model in step S300 is constructed based on a generalized regression neural network, and the hyperparameter is a smoothing factor in the generalized regression neural network.
在某些实施方式中，步骤S200中所述果蝇算法采用动态步长初始化群体中个体的位置。In some embodiments, the Drosophila algorithm described in step S200 uses a dynamic step size to initialize the positions of individuals in the population.
在某些实施方式中，步骤S200中所述果蝇算法采用交叉验证法，将种群分为多个相等的子种群，然后分别进行优化分析，最终再选择最优解。In some embodiments, the Drosophila algorithm described in step S200 adopts a cross-validation method, divides the population into multiple equal sub-populations, and then performs optimization analysis respectively, and finally selects the optimal solution.
在某些实施方式中，步骤S200中采用误差反向传播算法对预测模型进行训练。In some embodiments, the error back-propagation algorithm is used to train the prediction model in step S200.
在某些实施方式中，步骤S100中变压器状态参量的相关数据包括参量本身的数据和/或参量间的比值数据。In some embodiments, the relevant data of the transformer state parameters in step S100 includes the data of the parameters themselves and/or the ratio data between the parameters.
在某些实施方式中，步骤S100中变压器状态参量包括变压器本体参量和变电站环境参量。In some embodiments, the transformer state parameters in step S100 include transformer body parameters and substation environment parameters.
其中，在一些实施方式中，变压器本体参量包括变压器油中溶解的气体含量和/或变压器油温；在一些实施方式中，变电站环境参量包括气温、地面湿度、相对湿度、平均风速中的一种或多种。Wherein, in some embodiments, the parameters of the transformer body include the dissolved gas content in the transformer oil and/or the temperature of the transformer oil; in some embodiments, the environmental parameters of the substation include one of air temperature, ground humidity, relative humidity, and average wind speed or more.
其中，在一些实施方式中，变压器油中溶解的气体包括H2、CO、CH4、C2H2、 C2H4、C2H6中的一种或多种。Wherein, in some embodiments, the gas dissolved in the transformer oil includes one or more of H2 , CO, CH4 , C2 H2 , C2 H4 , and C2 H6 .
其中，在一些实施方式中，基于变压器油中溶解的气体，参量间的比值数据包括CH4/H2、C2H2/C2H4、C2H4/C2H6、C2H6/CH4、C2H2/CH4、C2H6/C2H2中的一种或多种。Wherein, in some embodiments, based on the gas dissolved in the transformer oil, the ratio data between parameters includes CH4 /H2 , C2 H2 /C2 H4 , C2 H4 /C2 H6 , C2 One or more of H6 /CH4 , C2 H2 /CH4 , C2 H6 /C2 H2 .
图2示意了基于果蝇算法优化的变压器状态参量数据预测方法在一种实施方式下的流程。FIG. 2 illustrates the flow of the transformer state parameter data prediction method optimized based on the Drosophila algorithm in one embodiment.
如图2所示，本发明所述的基于果蝇算法优化的变压器状态参量数据预测方法在一种实施方式下的流程包括以下步骤1-步骤6：As shown in FIG. 2 , the flow of the transformer state parameter data prediction method based on fruit fly algorithm optimization according to an embodiment of the present invention includes the following steps 1 to 6:
步骤1：通过数据采集模块获取一段时间内的变压器在线监测状态量数据。Step 1: Obtain the transformer online monitoring state quantity data for a period of time through the data acquisition module.
该步骤通过数据采集模块采集一段时间内的变压器在线监测状态量数据，其中，变压器在线监测状态量包括变压器状态参量及参量间比值。其中，变压器状态参量具体包括变压器油中溶解的气体含量、变压器油温等变压器本体参量，以及气温、地面温度、相对湿度和平均风速等变电站环境参量。其中，变压器油中溶解的气体包括氢气(H2)、一氧化碳(CO)、甲烷(CH4)、乙炔(C2H2)、乙烯(C2H4)以及乙烷(C2H6)，参量间比值采用6组变压器油中溶解的气体含量比值，以分子式之比的形式表示分别为CH4/H2，C2H2/C2H4， C2H4/C2H6，C2H6/CH4，C2H2/CH4，C2H6/C2H2。In this step, the data acquisition module collects the transformer online monitoring state quantity data within a period of time, wherein the transformer online monitoring state quantity includes the transformer state parameter and the ratio between the parameters. Among them, the transformer state parameters specifically include transformer body parameters such as the dissolved gas content in the transformer oil, transformer oil temperature, and substation environmental parameters such as air temperature, ground temperature, relative humidity, and average wind speed. Among them, the gas dissolved in the transformer oil includes hydrogen (H2 ), carbon monoxide (CO), methane (CH4 ), acetylene (C2 H2 ), ethylene (C2 H4 ) and ethane (C2 H6 ) , the ratio between parameters adopts the ratio of the dissolved gas content in 6 groups of transformer oil, expressed in the form of the ratio of molecular formula, respectively CH4 /H2 , C2 H2 /C2 H4 , C2 H4 /C2 H6 , C2 H6 /CH4 , C2 H2 /CH4 , C2 H6 /C2 H2 .
步骤2：通过数据处理模块将变压器在线监测状态量数据转换为变压器在线监测状态量矩阵X。Step 2: Convert the transformer online monitoring state quantity data into the transformer online monitoring state quantity matrix X through the data processing module.
该步骤通过数据处理模块采用离差标准化方法对变压器在线监测状态量数据进行归一化处理，得到变压器在线监测状态量矩阵X：In this step, the data processing module adopts the dispersion normalization method to normalize the transformer online monitoring state quantity data, and obtains the transformer online monitoring state quantity matrix X:
其中，X1、X2和Xr表示各变压器在线监测状态量，下标1,2……n表示时间序列。Among them, X1 , X2 and Xr represent the on-line monitoring state quantities of each transformer, and the subscripts 1, 2...n represent the time series.
步骤3：通过数据处理模块构建变压器状态参量数据预测模型。Step 3: Build a transformer state parameter data prediction model through the data processing module.
该步骤中，基于广义回归神经网络构建变压器状态参量数据预测模型，并通过多次反复试验选取最优结果来确定预测模型的结构。其中，预测模型的结构参数包括广义回归网络的隐藏层层数、神经元节点数。In this step, the transformer state parameter data prediction model is constructed based on the generalized regression neural network, and the optimal result is selected through repeated trials to determine the structure of the prediction model. Among them, the structural parameters of the prediction model include the number of hidden layers and the number of neuron nodes of the generalized regression network.
广义回归神经网络由输入层、模式层、求和层和输出层构成。广义回归神经网络的基本思想是利用非线性回归计算得到的，其灵感来源于统计学科中的概率论的特点。对于一个网络，当用函数进行逼近时，其网络输出就是网络对于输入向量的回归结果。设f(x,y)是随机变量x和y的概率密度函数，其中x是p维的输入向量，y是对应的输出向量。已知X是随机变量x的一组测量值，则y相对于X的回归值为：A generalized regression neural network consists of an input layer, a pattern layer, a summation layer and an output layer. The basic idea of generalized regression neural network is calculated by nonlinear regression, and its inspiration comes from the characteristics of probability theory in statistics. For a network, when a function is used for approximation, the network output is the regression result of the network on the input vector. Let f(x,y) be the probability density function of random variables x and y, where x is the p-dimensional input vector and y is the corresponding output vector. Given that X is a set of measurements of a random variable x, the regression value of y with respect to X is:
也是在输入为X的条件下所对应的y的预测输出。 It is also the predicted output of y corresponding to the input X.
当概率密度函数f(X,y)未知时，就需用应用Parzen非参数估计的方法，利用采集到的样本数据估计密度函数When the probability density function f(X, y) is unknown, it is necessary to use the method of applying Parzen non-parametric estimation, using the collected sample data Estimated density function
式中xi，yi是随机变量x和y的样本观测值；n是样本的容量；σ是高斯函数的宽度系数，也称光滑因子或者扩展参数。in the formula xi , yi are the sample observations of random variables x and y; n is the capacity of the sample; σ is the width coefficient of the Gaussian function, also known as the smoothing factor or expansion parameter.
将式(2)的代替(1)的f(X,y)，同时交换积分与加和的顺序，可得：The formula (2) will be Instead of f(X, y) in (1), and swap the order of integration and summation at the same time, we can get:
因为所以式(3)以进一步化简为：because So formula (3) is further simplified as:
由式(4)可知，构建GRNN网络时，网络需要训练的超参数只有一个参数σ需要设定。当光滑因子σ趋向于无穷大时，d(X,xi)趋向于零。从而代表所有样本因变量的均值。当光滑因子σ趋向于零时，则和训练样本非常接近。此时如果预测的点在训练样本集合中时，预测值会和样本的期望值十分接近，但是一旦是新的数据，即是样本中未能包含进去的点，则预测效果可能会非常差，无法满足预测要求。所以只有当光滑因子σ取值合适时，所有的训练样本的因变量才能被充分考虑进去，与预测点距离近的样本点所对应的因变量也被附加了更大的权重。传统的方式是通过交叉验证法来得到光滑因子σ，在一定程度上提高了网络的逼近能力和分类能力。本发明则在下面的步骤4中结合改进的果蝇优化算法来选取光滑因子σ，算例结果表明该方法具有更好的全局收敛性，在预测过程中提高了预测的精度和可靠性。It can be seen from equation (4) that when building a GRNN network, only one hyperparameter σ needs to be set for the network's hyperparameters to be trained. As the smoothing factor σ tends to infinity, d(X, xi ) tends to zero. thereby represents the mean of all sample dependent variables. When the smoothing factor σ tends to zero, is very close to the training sample. At this time, if the predicted point is in the training sample set, the predicted value will be very close to the expected value of the sample, but once it is new data, that is, the point that cannot be included in the sample, the prediction effect may be very poor. meet forecast requirements. Therefore, only when the value of the smoothing factor σ is appropriate, the dependent variables of all training samples can be fully considered, and the dependent variables corresponding to the sample points that are close to the prediction point are also given greater weights. The traditional method is to obtain the smooth factor σ through the cross-validation method, which improves the approximation ability and classification ability of the network to a certain extent. The present invention selects the smoothing factor σ in combination with the improved fruit fly optimization algorithm in the following step 4. The calculation example results show that the method has better global convergence, and improves the accuracy and reliability of the prediction during the prediction process.
步骤4：通过数据处理模块基于果蝇算法求得预测模型的光滑因子。Step 4: Obtain the smooth factor of the prediction model based on the Drosophila algorithm through the data processing module.
果蝇算法步骤如下：The steps of the fruit fly algorithm are as follows:
步骤4-1：在每一次迭代中，光滑因子和模型评判指标均方根误差值作为一对参数，对其进行随机初始化Init_X，Init_Y。Step 4-1: In each iteration, the smoothing factor and the root mean square error value of the model evaluation index are used as a pair of parameters to randomly initialize Init_X, Init_Y.
步骤4-2：初始化群体中个体的位置：Step 4-2: Initialize the positions of individuals in the population:
xi＝Init_X+l*r (5)xi =Init_X+l*r (5)
yi＝Init_Y+l*r (6)yi =Init_Y+l*r (6)
式中，l表示步长值，r是[0,1]区间的随机数，xi,yi为个体位置坐标。In the formula, l represents the step value, r is a random number in the [0,1] interval, and xi , yi are the individual position coordinates.
步骤4-3：计算个体与原点的距离di，并求判定值pi.。Step 4-3: Calculate the distance di between the individual and the origin, and find the judgment value pi. .
pi＝1/di (8)pi =1/di (8)
步骤4-4：找出群体中最优个体。Step 4-4: Find the best individual in the group.
[bestS,bestIndex]＝max(p) (9)[bestS,bestIndex]=max(p) (9)
步骤4-5：更新判定值和新的坐标。Step 4-5: Update the decision value and the new coordinates.
bestSmell＝bestS (10)bestSmell=bestS(10)
x＝x(bestIndex) (11)x=x(bestIndex) (11)
y＝y(bestIndex) (12)y=y(bestIndex) (12)
式中bestSmell即为模型的最优光滑因子参数。where bestSmell is the optimal smooth factor parameter of the model.
由于在步骤4-2中，传统方法的步长是固定的。如果设定的步长过大，则会导致算法的搜索能力变弱，花费的搜索时间过长，导致算法的效率偏低；如果设置的步长过小，则算法容易陷入局部最优。其次传统的果蝇算法存在收敛精度低的问题。本发明对步长值的选取进行改进，采用动态步长来代替固定步长。优化的步长公式如下：Since in step 4-2, the step size of the traditional method is fixed. If the set step size is too large, the search ability of the algorithm will be weakened, and the search time will be too long, resulting in low efficiency of the algorithm; if the set step size is too small, the algorithm will easily fall into a local optimum. Secondly, the traditional fruit fly algorithm has the problem of low convergence accuracy. The invention improves the selection of the step size, and adopts the dynamic step size to replace the fixed step size. The optimized step size formula is as follows:
式中，ki是当前的迭代次数，k是最大迭代次数，l0是初始步长，li是当前步长。In the formula,ki is the current number of iterations,k is the maximum number of iterations, l0 is the initial step size, and li is the current step size.
而针对收敛精度低的问题，本发明采用了交叉验证的方法，将种群分为多个相等的子种群，然后分别进行优化分析，最终再选择最好的解。该方案保证了算法能够充分利用数据，防止算法陷入局部最优解。For the problem of low convergence precision, the present invention adopts the method of cross-validation, divides the population into a plurality of equal sub-populations, and then performs optimization analysis respectively, and finally selects the best solution. This scheme ensures that the algorithm can make full use of the data and prevents the algorithm from falling into a local optimal solution.
步骤5：通过数据处理模块基于变压器在线监测状态量矩阵X对预测模型进行训练。Step 5: Train the prediction model based on the online monitoring state quantity matrix X of the transformer through the data processing module.
该步骤中，将变压器在线监测状态量矩阵X中的变压器在线监测状态量前80％的数据作为输入，采用误差反向传播算法对预测模型进行训练，确定该预测模型的结构参数。In this step, the data of the top 80% of the transformer online monitoring state quantities in the transformer online monitoring state quantity matrix X are used as input, and the error back propagation algorithm is used to train the prediction model, and the structural parameters of the prediction model are determined.
步骤6：通过数据处理模块基于经训练的变压器状态参量数据预测模型预测变压器状态参量数据。Step 6: Predict the transformer state parameter data based on the trained transformer state parameter data prediction model through the data processing module.
该步骤中，将变压器在线监测状态量矩阵X中的剩下的20％的变压器在线监测状态量作为输入数据输入经过训练的预测模型，输入给GRNN神经网络层，神经网络层输出预测结果。In this step, the remaining 20% of the transformer online monitoring state quantities in the transformer online monitoring state quantity matrix X are input into the trained prediction model as input data, and input to the GRNN neural network layer, and the neural network layer outputs the prediction result.
上述数据采集模块和数据处理模块相互数据连接，构成本实施例的基于果蝇算法优化的变压器状态参量数据预测系统。该系统采用上述变压器状态参量数据预测方法对变压器状态参量数据进行预测。The above-mentioned data acquisition module and data processing module are connected to each other in data, and constitute the transformer state parameter data prediction system optimized based on the fruit fly algorithm of this embodiment. The system uses the above-mentioned transformer state parameter data prediction method to predict the transformer state parameter data.
下面对上述实施例进行测试验证。The above embodiments are tested and verified below.
图3示意了本验证实例中光滑因子优化过程的均方根误差Rmse变化曲线，Figure 3 shows the variation curve of the root mean square error Rmse of the smooth factor optimization process in this verification example,
图4示意了本验证实例中甲烷气体浓度预测结果相对百分误差。Figure 4 illustrates the relative percentage error of the methane gas concentration prediction results in this verification example.
本验证实例采用上述实施例的变压器状态参量数据预测方法和系统对变压器状态参量数据进行预测。基于某220kV变压器油色谱在线监测装置油色谱数据获取变压器在线监测状态量数据，数据的采样间隔为1天。将400组监测数据作为训练样本，将30组监测数据作为测试样本。In this verification example, the transformer state parameter data prediction method and system of the above embodiments are used to predict the transformer state parameter data. Based on the oil chromatography data of a 220kV transformer oil chromatography online monitoring device, the transformer online monitoring state quantity data was obtained, and the data sampling interval was 1 day. 400 sets of monitoring data were used as training samples, and 30 sets of monitoring data were used as test samples.
为了评价本发明提出的组合预测模型的准确性和有效性，采用以下几个评价准则来进行分析：In order to evaluate the accuracy and validity of the combined prediction model proposed by the present invention, the following evaluation criteria are used for analysis:
测试集的真实值和预测值的均方根误差值Rmse，其表达式为：The root mean square error value Rmse of the true value and the predicted value of the test set is expressed as:
真实值和预测值的平均相对百分误差，其表达式为：The average relative percent error between the true value and the predicted value, which is expressed as:
最大相对百分误差，其表达式为：The maximum relative percent error is expressed as:
式中，N是测试集数据的个数，xi是真实值，是预测值。In the formula, N is the number of test set data, xi is the real value, is the predicted value.
本验证实例以甲烷CH4气体浓度预测结果为例说明整个预测的流程。首先输入的变压器在线监测状态量向量依次为：CH4(仅历史浓度)、H2、CO、C2H2、C2H4、C2H6、总烃、总可燃气浓度、C2H2/C2H4、C2H4/C2H6、环境温度和油温共11 组监测数据。将这些向量进行标准化处理，映射到[0,1]之间，转换函数为：This verification example takes the prediction result of methane CH4 gas concentration as an example to illustrate the whole prediction process. The first input transformer online monitoring state quantity vector is: CH4 (only historical concentration), H2 , CO, C2 H2 , C2 H4 , C2 H6 , total hydrocarbons, total combustible gas concentration, C2 H2 /C2 H4 , C2 H4 /C2 H6 , ambient temperature and oil temperature, a total of 11 sets of monitoring data. These vectors are normalized and mapped to [0,1], and the conversion function is:
式中，xmin为向量样本数据的最小值，xmax为向量样本数据的最大值。In the formula, xmin is the minimum value of the vector sample data, and xmax is the maximum value of the vector sample data.
基于广义回归神经网络构建的变压器状态参量数据预测模型。Transformer state parameter data prediction model based on generalized regression neural network.
基于改进的果蝇算法确定预测模型的光滑因子σ。光滑因子优化过程的均方根误差Rmse变化曲线如图3所示，通过100次迭代，最终在第44次迭代中收敛，此时的Rmse值最小为0.0105，对应的光滑因子σ的值为0.0867。The smoothing factor σ of the prediction model is determined based on the improved Drosophila algorithm. The root mean square error Rmse variation curve of the smooth factor optimization process is shown in Figure 3. After 100 iterations, it finally converges in the 44th iteration. At this time, the minimum value of Rmse is 0.0105, and the corresponding smooth factor σ value is 0.0867.
基于上述向量训练基于广义回归神经网络构建的变压器状态参量数据预测模型，确定模型的隐藏层层数、神经元节点数。Based on the above vector training, the transformer state parameter data prediction model constructed based on the generalized regression neural network is trained, and the number of hidden layers and the number of neuron nodes of the model is determined.
基于经训练的变压器状态参量数据预测模型对甲烷气体浓度进行预测，真实值和预测值的相对百分误差如图4所示，横坐标为时间序列，单位为天，纵坐标为相对百分误差，单位为％，图中平均相对百分误差为0.55％，最大相对百分误差1.32％。The methane gas concentration is predicted based on the trained transformer state parameter data prediction model. The relative percentage error between the actual value and the predicted value is shown in Figure 4. The abscissa is the time series, the unit is days, and the ordinate is the relative percentage error. , the unit is %, the average relative percentage error in the figure is 0.55%, and the maximum relative percentage error is 1.32%.
将本发明的预测方法同传统的支持向量机(SVM)、反馈神经网络(BPNN) 预测方法的结果进行比较。预测的变压器状态参量数据为甲烷气体浓度数据，测试集的个数为30，预测效果如表1所示。The prediction method of the present invention is compared with the results of the traditional support vector machine (SVM) and feedback neural network (BPNN) prediction methods. The predicted transformer state parameter data is methane gas concentration data, the number of test sets is 30, and the prediction effect is shown in Table 1.
表1.不同预测方法对预测效果的影响Table 1. Effects of different forecasting methods on forecasting performance
从表1可以看出，在测试集的样本数相同的情况下，本发明提出的基于果蝇算法优化的变压器状态参量数据预测方法和系统预测结果更为准确，对数据的拟合较好。It can be seen from Table 1 that under the condition of the same number of samples in the test set, the transformer state parameter data prediction method and system prediction result based on the fruit fly algorithm optimization proposed by the present invention are more accurate and fit the data better.
不同的测试集数目对预测的准确性也会造成一定的影响。分别对10、30、 60、90个数据点进行预测，预测结果如表2所示。Different number of test sets will also have a certain impact on the accuracy of prediction. 10, 30, 60, and 90 data points are predicted respectively, and the prediction results are shown in Table 2.
表2.不同测试集样本数目对预测效果的影响Table 2. The effect of the number of samples in different test sets on the prediction effect
从表2中可以得出：当测试集样本数目较小时，本发明提出的预测模型比其它几种模型预测的准确率要高。随着测试集样本数目的增多，几种模型的预测准确度都有所下降，但本发明提出的预测方法在预测性能上明显优于其余两种模型。It can be concluded from Table 2 that when the number of samples in the test set is small, the prediction model proposed by the present invention has a higher prediction accuracy than other models. As the number of test set samples increases, the prediction accuracy of several models decreases, but the prediction method proposed in the present invention is obviously better than the other two models in prediction performance.
要注意的是，以上列举的仅为本发明的具体实施例，显然本发明不限于以上实施例，随之有着许多的类似变化。本领域的技术人员如果从本发明公开的内容直接导出或联想到的所有变形，均应属于本发明的保护范围。It should be noted that the above enumeration is only a specific embodiment of the present invention, and it is obvious that the present invention is not limited to the above embodiment, and there are many similar changes. All modifications directly derived or thought of by those skilled in the art from the content disclosed in the present invention shall belong to the protection scope of the present invention.
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910276198.9ACN110045237A (en) | 2019-04-08 | 2019-04-08 | Transformer state parametric data prediction technique and system based on drosophila algorithm optimization |
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910276198.9ACN110045237A (en) | 2019-04-08 | 2019-04-08 | Transformer state parametric data prediction technique and system based on drosophila algorithm optimization |
| Publication Number | Publication Date |
|---|---|
| CN110045237Atrue CN110045237A (en) | 2019-07-23 |
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910276198.9APendingCN110045237A (en) | 2019-04-08 | 2019-04-08 | Transformer state parametric data prediction technique and system based on drosophila algorithm optimization |
| Country | Link |
|---|---|
| CN (1) | CN110045237A (en) |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110674604A (en)* | 2019-09-20 | 2020-01-10 | 武汉大学 | Transformer DGA data prediction method based on multi-dimensional time sequence frame convolution LSTM |
| CN111651936A (en)* | 2020-05-27 | 2020-09-11 | 辽宁工程技术大学 | A FOA-GRNN-based modeling and design method for dual-notch ultra-wideband antennas |
| CN113393107A (en)* | 2021-06-07 | 2021-09-14 | 东方电气集团科学技术研究院有限公司 | Incremental calculation method for state parameter reference value of power generation equipment |
| CN113435486A (en)* | 2021-06-15 | 2021-09-24 | 西安科技大学 | Coal gangue identification method based on PCA-IFOA-SVM combined with gray level-texture fusion features |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103942625A (en)* | 2014-04-23 | 2014-07-23 | 天津大学 | Hydropower station plant dam structure vibration response prediction method based on FOA-GRNN |
| CN107677473A (en)* | 2017-09-23 | 2018-02-09 | 哈尔滨理工大学 | A kind of GRNN rotating machinery fault Forecasting Methodologies based on FOA optimizations |
| CN108390380A (en)* | 2018-05-16 | 2018-08-10 | 国网山东省电力公司青岛供电公司 | A kind of transformer state parameter trend forecasting method and system |
| CN108665052A (en)* | 2017-03-30 | 2018-10-16 | 华为技术有限公司 | The method and apparatus of drosophila algorithm optimization generalized regression nerve networks |
| CN109164248A (en)* | 2018-09-18 | 2019-01-08 | 山东理工大学 | A kind of predicting model for dissolved gas in transformer oil method |
| CN109242205A (en)* | 2018-10-09 | 2019-01-18 | 国网上海市电力公司 | A kind of transformer state prediction technique and system |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103942625A (en)* | 2014-04-23 | 2014-07-23 | 天津大学 | Hydropower station plant dam structure vibration response prediction method based on FOA-GRNN |
| CN108665052A (en)* | 2017-03-30 | 2018-10-16 | 华为技术有限公司 | The method and apparatus of drosophila algorithm optimization generalized regression nerve networks |
| CN107677473A (en)* | 2017-09-23 | 2018-02-09 | 哈尔滨理工大学 | A kind of GRNN rotating machinery fault Forecasting Methodologies based on FOA optimizations |
| CN108390380A (en)* | 2018-05-16 | 2018-08-10 | 国网山东省电力公司青岛供电公司 | A kind of transformer state parameter trend forecasting method and system |
| CN109164248A (en)* | 2018-09-18 | 2019-01-08 | 山东理工大学 | A kind of predicting model for dissolved gas in transformer oil method |
| CN109242205A (en)* | 2018-10-09 | 2019-01-18 | 国网上海市电力公司 | A kind of transformer state prediction technique and system |
| Title |
|---|
| 朱沛恒: "基于果蝇算法优化的概率神经网络在变压器故障诊断中的应用", 《电力大数据》* |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110674604A (en)* | 2019-09-20 | 2020-01-10 | 武汉大学 | Transformer DGA data prediction method based on multi-dimensional time sequence frame convolution LSTM |
| CN110674604B (en)* | 2019-09-20 | 2022-07-08 | 武汉大学 | Transformer DGA data prediction method based on multi-dimensional time series frame convolution LSTM |
| CN111651936A (en)* | 2020-05-27 | 2020-09-11 | 辽宁工程技术大学 | A FOA-GRNN-based modeling and design method for dual-notch ultra-wideband antennas |
| CN113393107A (en)* | 2021-06-07 | 2021-09-14 | 东方电气集团科学技术研究院有限公司 | Incremental calculation method for state parameter reference value of power generation equipment |
| CN113393107B (en)* | 2021-06-07 | 2022-08-12 | 东方电气集团科学技术研究院有限公司 | An Incremental Calculation Method for the Reference Value of State Parameters of Power Generation Equipment |
| CN113435486A (en)* | 2021-06-15 | 2021-09-24 | 西安科技大学 | Coal gangue identification method based on PCA-IFOA-SVM combined with gray level-texture fusion features |
| Publication | Publication Date | Title |
|---|---|---|
| Cheng et al. | A hybrid remaining useful life prognostic method for proton exchange membrane fuel cell | |
| CN105117602B (en) | A kind of metering device running status method for early warning | |
| CN108038300B (en) | Optical fiber state assessment method based on improved membership function combined with neural network | |
| CN106920007B (en) | PM based on second-order self-organizing fuzzy neural network2.5Intelligent prediction method | |
| CN110348615B (en) | Cable line fault probability prediction method based on ant colony optimization support vector machine | |
| CN111426816A (en) | Method for predicting concentration of dissolved gas in transformer oil based on PSO-L STM | |
| CN110045237A (en) | Transformer state parametric data prediction technique and system based on drosophila algorithm optimization | |
| CN105631578A (en) | Risk assessment-orientated modeling method of power transmission and transformation equipment failure probability model | |
| CN116204794B (en) | Method and system for predicting dissolved gas in transformer oil by considering multidimensional data | |
| CN110598854A (en) | GRU model-based transformer area line loss rate prediction method | |
| CN107274067B (en) | A risk assessment method for distribution transformer overload | |
| CN112685961A (en) | Method and system for predicting remaining service life of analog circuit | |
| CN106295798A (en) | Empirical mode decomposition and Elman neural network ensemble wind-powered electricity generation Forecasting Methodology | |
| CN104700321A (en) | Analytical method of state running tendency of transmission and distribution equipment | |
| CN110441500A (en) | A kind of predicting model for dissolved gas in transformer oil method based on shot and long term memory network | |
| CN107609671A (en) | A kind of Short-Term Load Forecasting Method based on composite factor evaluation model | |
| CN113782113B (en) | Method for identifying gas fault in transformer oil based on deep residual error network | |
| CN113591402A (en) | Digital power transformer health state fuzzy comprehensive evaluation method | |
| CN118188342A (en) | Fan-oriented fault early warning and life prediction method | |
| CN113379116A (en) | Cluster and convolutional neural network-based line loss prediction method for transformer area | |
| CN108090615A (en) | Low-limit frequency Forecasting Methodology after electric power system fault based on cross entropy integrated study | |
| CN117522066A (en) | Combined optimization method and system based on peak shaving power supply equipment combination prediction | |
| CN116129212A (en) | Power load prediction method and system based on graph neural network | |
| CN113469409A (en) | Gaussian process-based state prediction method and device for electric energy metering device | |
| CN110045236A (en) | Transformer state parametric data prediction technique and system based on core pivot element analysis optimization |
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication | Application publication date:20190723 | |
| WD01 | Invention patent application deemed withdrawn after publication |