A method for phase-loss fault diagnosis in distribution network based on DBN and rough set neural network

Abstract

The present invention discloses a distribution network phase-loss fault diagnosis method based on DBN and rough set neural network, including: obtaining sample data such as the incoming line voltage parameters, outgoing line voltage and current parameters, and air temperature in the JP cabinet of the load distribution switch collected by the terminal detection equipment in the JP cabinet; performing data preprocessing on the collected data, and inputting the sample data into the deep belief network for training; extracting features and parameters from the output labels of the deep belief network; importing the features and parameter labels into the rough set model for simplification; using RBF neural network for learning training and fault diagnosis, and outputting the distribution network phase-loss fault diagnosis result. The present invention combines the deep belief network and the rough neural network to diagnose the distribution network phase-loss fault, has good parameter extraction ability and generalization ability, has good fault tolerance, solves the problem of low recognition rate of distribution network phase-loss working condition fault, and realizes accurate and rapid diagnosis of distribution network phase-loss fault.

Description

Technical Field

[0001] The invention belongs to the field of fault diagnosis, and in particular relates to a distribution network phase-loss fault diagnosis method based on DBN and rough set neural network. Technical Background

[0002] The distribution network is an important part of the power system. It plays the role of distributing electric energy. Its power supply reliability and quality directly affect social production and economic development, and are also closely related to people's daily lives. Due to the wide distribution of the distribution network, complex structure, large differences in urban and rural construction levels, unbalanced maintenance efforts, and harsh line environment in mountainous areas, distribution line failures occur frequently, seriously threatening the safe and stable operation of the power system.

[0003] In the distribution network, the voltage of a certain phase increases due to phase loss, zero break fault and load imbalance in the distribution line, which often damages the electrical equipment and causes serious economic losses. In theory, it is difficult to distinguish between phase loss and no-load conditions, and zero break conditions and no-load and three-phase imbalance conditions. There are no definitions and technical requirements for phase loss and zero break faults in the current national standards for low-voltage electrical appliances, and there are few practical phase loss and zero break fault diagnosis and online monitoring products in China. With the development of the requirements for intelligent and digital construction of power grids, it is of great practical significance to study a method for monitoring and diagnosing phase loss faults in distribution networks that can accurately locate, is low-cost and easy to monitor quickly over a wide area, which is of great practical significance for improving the power supply reliability of distribution networks and the service capabilities of power supply units.

[0004] At present, the fault diagnosis method of distribution network is mainly realized through the action information of switch elements in the line or the information of fault recorder. The main methods include expert system, artificial neural network, Bayesian network, etc. These methods have problems such as slow speed, poor fault tolerance, large diagnostic error, and poor autonomous learning ability. Deep belief network (DBN) has many advantages such as simple model structure, low training difficulty, fast convergence speed, and higher feature extraction and generalization ability compared with traditional networks. Rough set theory can well handle the situation of incomplete information and information redundancy, and is suitable for the diagnosis of complex fault conditions in distribution network. However, there is no distribution network phase failure fault diagnosis method based on deep belief network and rough set neural network.

[0005] The purpose of the present invention is to solve the shortcomings of the existing technologies and methods for the diagnosis of distribution network phase loss faults, and to provide a distribution network phase loss fault diagnosis method based on DBN and rough set neural network. Combining the advantages of deep belief network deep learning algorithm and rough neural network algorithm, the model is trained according to historical fault information to obtain the distribution network phase loss fault characteristics, so as to accurately and quickly diagnose the distribution network phase loss fault. Summary of the invention

[0006] The technical solution of the present invention is a distribution network phase failure diagnosis method based on a deep belief network, which is characterized by comprising the following steps:

[0007] Step S11 obtains sample data such as the incoming line voltage parameters, outgoing line voltage and current parameters, and air temperature in the JP cabinet of the load distribution switch collected by the terminal detection equipment in the JP cabinet;

[0008] Step S12 performs data preprocessing on the collected data, inputs the sample data into the deep belief network, and uses the deep learning algorithm model for training;

[0009] Step S13 extracts features and parameters from the output labels of the deep belief network;

[0010] Step S14: import the feature quantity and parameter label into the rough set model for simplification;

[0011] Step S15 uses the RBF neural network for learning, training and fault diagnosis, and outputs the distribution network phase loss fault diagnosis result.

[0012] Compared with the prior art, the present invention has the following significant advantages: 1) The present invention uses a genetic algorithm based on rough set theory to simplify the original fault decision table, and uses the simplified decision table as a learning sample of the RBF neural network, which can solve the problem of redundant monitoring data features and uncertain information in fault diagnosis, and can simplify the data and obtain the minimum expression of knowledge while retaining key information, revealing a simple pattern of information sets and improving the efficiency of distribution network phase failure diagnosis; 2) The present invention uses a deep belief network to process voltage, current and temperature signals, and realizes the conversion of input data through a multi-layer RBM structure, thereby improving the analysis capability of real-time fault feedback information, having higher eigenvalue and parameter extraction capabilities and generalization capabilities, improving fault detection accuracy, and increasing convergence speed. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the technical solution of the present invention, the drawings required for use in the embodiments are briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0014] Figure 1 A schematic diagram of a process of an embodiment of the present invention

[0015] Figure 2 Schematic diagram of the fault diagnosis method according to an embodiment of the present invention

[0016] Figure 3 Fault judgment calculation conditions for RBF neural network training and learning DETAILED DESCRIPTION

[0017] In order to solve the problem of diagnosing and monitoring the phase loss fault of the distribution network, the present invention discloses a distribution network phase loss fault diagnosis method based on DBN and rough set neural network through the following embodiments.

[0018] The embodiment of the present invention discloses a method for diagnosing phase failure in a distribution network based on DBN and rough set neural network, and its principle diagram is shown in FIG. Figure 1 As shown, including:

[0019] Step S11 obtains sample data such as the incoming line voltage parameters, outgoing line voltage and current parameters, and air temperature in the JP cabinet of the load distribution switch collected by the terminal detection equipment in the JP cabinet;

[0020] Step S12 performs data preprocessing on the collected data, inputs the sample data into DBN, and uses the DBN deep learning algorithm model for training;

[0021] Step S13 extracts features and parameters from the output labels of the deep belief network;

[0022] Step S14: import the feature quantity and parameter label into the rough set model for simplification;

[0023] Step S15 uses the RBF neural network for learning, training and fault diagnosis, and outputs the distribution network phase loss fault diagnosis result.

[0024] The following is combined with Figure 2 This embodiment is described in detail:

[0025] Step S11, obtaining sample data such as the incoming voltage parameter, outgoing voltage and current parameter, and air temperature in the JP cabinet of the load distribution switch collected by the terminal detection device in the JP cabinet. The incoming voltage parameter of the load distribution switch in the JP cabinet is the measured value of the three-phase voltages A, B, and C on the incoming side of the load distribution switch in the JP cabinet; the outgoing voltage and current parameter of the JP cabinet is the measured value of the three-phase voltages A, B, and C on the outgoing side of the load distribution switch.

[0026] Step S12, preprocess the collected data, input the sample data into DBN, and use the DBN deep learning algorithm model for training. The DBN deep learning algorithm model is a 5-layer DBN structure, consisting of 1 input layer, 1 label layer and 3 hidden layers, and the layers are fully connected. Among them, ν is the input feature vector of the input layer, which is the digital signal converted by the analog-to-digital converter of the detection signals such as the three-phase bus voltage on the secondary side of the distribution network transformer or the three-phase voltage of the incoming line of the JP cabinet of the distribution network, the three-phase current and zero-phase current on the outgoing line side of each load distribution switch of the JP cabinet, and the ambient temperature measurement value; h c,jis the value of the jth neuron in the cth hidden layer, J is the number of nodes including neurons in each layer; l is the output label vector, ω is the weight matrix between the visible and hidden layers, which represents the connection weight between the ith neuron in the visible layer and the jth neuron in the hidden layer, and a and b represent the bias vectors of the visible and hidden layers respectively.

[0027] The DBN deep learning algorithm uses two key steps for training and learning: unsupervised training based on the RBM contrast divergence algorithm and supervised fine-tuning based on the back-propagation algorithm.

[0028] The parameters are learned by layer-by-layer unsupervised domain training. First, the input feature vector ν and the first hidden layer h1 are taken as an RBM1, and the parameters of RBM1 (ω1, b1) are trained. Then, the parameters of RBM1 are fixed, h1 is regarded as a visible vector, and h2 is regarded as a hidden vector. RBM2 is trained to obtain its parameters (ω2, b2). Then, these parameters are fixed and RBM3 is trained.

[0029] Explicit layer node vector ν I and hidden layer node vector h J The joint state energy function is:

[0030]

[0031] In the formula, I and J are the number of nodes in the visible and hidden layers respectively;

[0032] The a and b vectors are:

[0033] a=(a 1 ,a 1 ,…,a i ,…,a I )

[0034] b=(b 1 ,b 1 ,…,b j ,…,b J )

[0035] The activation probability of the i-th explicit layer node is:

[0036]

[0037] The activation probability of the jth hidden layer node is:

[0038]

[0039] The DBN training and tuning process is as follows:

[0040] The first step is to initialize the learning rate η, the weight ω and offset a, b of the RBM structure parameters, so that (ω, a, b) are small random numbers, and the weight and offset change values ​​are set to 0, that is, Δω = 0, Δa = 0, Δb = 0;

[0041] In the second step, the fault data training sample is input, the CD-k algorithm is used to calculate the visible layer neuron vector, and Δω, Δa and Δb are updated;

[0042] Step 3: Update the RBM structure parameters (ω, a, b); repeat the second and third steps to train the RBM network until the set number of training times is reached or the learning rate reaches the set value;

[0043] The fourth step is to improve the classification accuracy of DBN through supervised reverse fine-tuning.

[0044] Step S13, extracting features and parameters from the output labels of the deep belief network.

[0045] The EMD-SVD feature extraction method is used to extract features and parameters from the output labels of the deep belief network. The steps are as follows:

[0046] In the first step, according to the EMD decomposition method, the DBN label vector signal is decomposed by EMD, and the intrinsic mode matrix is ​​formed using each simf after denoising;

[0047] The second step is to find the marginal spectrum of each IMF component, calculate the IMF marginal spectrum energy, and construct the eigenvector S with the IMF marginal spectrum energy as the element:

[0048] S=[S 1 ,S 1 ,…S n ]

[0049] In the third step, SVD is performed on the matrix, the distribution network fault type is used as the decision attribute, the elements in S are used as the conditional attributes, and the singular value vector is obtained as the feature vector to input into the rough set reduction model to obtain a simplified distribution network phase failure fault diagnosis decision table.

[0050] Step S14, importing the feature quantity and parameter label into the rough set model for simplification. The rough set neural network includes a rough set simplification layer model and an RBF neural network. The rough set layer model uses a genetic algorithm of rough set theory to simplify the original fault decision table, and uses the simplified decision table as a learning sample of the RBF neural network to solve the problem of redundant and uncertain information of monitoring data features in fault diagnosis. Under the premise of retaining key information, the data is simplified and the minimum expression of knowledge is obtained, revealing a simple pattern of information set.

[0051] Step S15, using the RBF neural network for learning training and fault diagnosis, and outputting the distribution network phase loss fault diagnosis result.

[0052] The rough set neural network includes a rough set reduction layer and a neural network. The neural network is a radial basis function RBF neural network. The RBF neural network includes an input layer, a hidden layer and an output layer. The rough set neural network includes n input samples, which become m samples as inputs of the RBF neural network after the rough set reduction layer. The number of inputs of the RBF neural network is m. The radial basis function adopts a hyperbolic tangent function, and the output layer adopts a linear function. The input and output responses of the RBF neural network are:

[0053]

[0054] In the formula, R i is the RBF neural network input vector, exp(·) is the radial basis function, ||·|| is the Euclidean space, d i is the network center, ψ ij is the connection weight between neurons in each layer of the RBF network, m is the number of hidden layer nodes; δ i is the width of the i-th basis function, y j is the actual output value of the jth output node of the network corresponding to the input sample.

[0055] The RBF neural network is used to judge the phase failure condition of the distribution network, such as Figure 3 :For each fundamental voltage amplitude U x (where x is the phase sequence number of phases A, B, and C), fundamental current amplitude I mx (where m = 1, 2, ..., is the serial number of the load distribution switch on the secondary side of the transformer, and x is the phase serial number of the three-phase four-wire system A, B, C, N) and the harmonic current content I hmx (where m is the serial number of the load distribution switch on the secondary side of the transformer, and x is the phase serial number of the three-phase four-wire system of A, B, C, and N) is used to determine: when U x Less than the threshold U L ,I mx Both are less than the threshold I L When the two conditions are met at the same time, the diagnosis result is that the transformer primary side x phase has a phase failure (where U L is a preset voltage threshold, generally between 50V and 110V, IL is a preset current threshold, generally between 2A and 5A, and the specific threshold is determined according to transformer parameters and measurement accuracy); when U x Greater than the threshold U L ,I mx Less than threshold I L ,I hmx Less than threshold IhL When the three conditions are met at the same time, the diagnosis result is that the x-phase under the load distribution switch m on the secondary side of the transformer has a phase failure (where U L It is a preset voltage threshold, generally between 50V and 110V. The specific threshold is determined according to the transformer parameters and measurement accuracy. hL It is a preset current threshold, generally between 0.5% and 1%, and the specific threshold is determined according to the local load conditions).

Claims

1. A method for diagnosing phase failure in distribution network based on DBN and rough set neural network. Features The following steps are involved: Step S11 obtains sample data of the incoming line voltage parameters, outgoing line voltage and current parameters, and air temperature in the JP cabinet of the load distribution switch collected by the terminal detection equipment in the JP cabinet; Step S12 performs data preprocessing on the collected data, inputs the sample data into the deep belief network, and uses the deep learning algorithm model for training; Step S13 extracts features and parameters from the output labels of the deep belief network; Step S14: import the feature quantity and parameter label into the rough set model for simplification; Step S15 uses the RBF neural network for learning, training and fault diagnosis, and outputs the distribution network phase loss fault diagnosis result.

2. The method according to claim 1, Features: The input feature vector of the input layer of the deep belief network is a digital signal converted by an analog-to-digital converter from the detection signals of the three-phase bus voltage on the secondary side of the distribution network transformer or the three-phase voltage on the incoming line of the JP cabinet of the distribution network, the three-phase current and zero-phase current on the outgoing line side of each load distribution switch of the JP cabinet, and the ambient temperature measurement value.

3. The method according to claim 1, Features: The characteristic quantities and parameter labels are introduced into the rough set model for simplification; the rough set neural network includes a rough set simplification layer model and an RBF neural network; the rough set simplification layer model uses a genetic algorithm of rough set theory to simplify the original fault decision table, and uses the simplified decision table as a learning sample of the RBF neural network to solve the problem of redundant monitoring data features and uncertain information in fault diagnosis.

4. The method according to claim 3, Features: The RBF neural network uses the following method to determine the phase failure condition of the distribution network: x , where x is the phase sequence number of phases A, B, and C, and the fundamental current amplitude I mx , where m = 1, 2, ..., is the serial number of the load distribution switch on the secondary side of the transformer, and x is the phase serial number of the three-phase four-wire system A, B, C, N and the harmonic current content I hmx , where m is the serial number of the load distribution switch on the secondary side of the transformer, and x is the characteristic value of the phase serial number of the three-phase four-wire system of A, B, C, and N. x Less than the threshold U L ,I mx Both are less than the threshold I L When both conditions are met, the diagnosis result is that the transformer primary side x phase has a phase failure, where U L is a preset voltage threshold between 50V and 110V, IL is a preset current threshold between 2A and 5A, and the specific threshold is determined according to the transformer parameters and measurement accuracy; when U x Greater than the threshold U L ,I mx Less than threshold I L ,I hmx Less than threshold I hL When the three conditions are met at the same time, the diagnosis result is that the x-phase under the load distribution switch m on the secondary side of the transformer has a phase failure, where U L is a preset voltage threshold between 50V and 110V. The specific threshold is determined according to the transformer parameters and measurement accuracy. hL It is a preset current threshold value between 0.5% and 1%, and the specific threshold value is determined according to the local load conditions.

Images