Power transmission line fault reason identification method based on experience guidance

Abstract

The invention discloses a power transmission line fault reason identification method based on experience guidance, and belongs to the technical field of power transmission line fault diagnosis and artificial intelligence deep learning. Expanding the unbalanced sample through a synthetic minority class oversampling technology; implicit features in fault transient waveforms are fully extracted through a long short-term memory neural network, and a nonlinear mapping relation between fault recording and fault reasons is preliminarily established; and the engineering experience of the operation and maintenance personnel is sorted, and the expert knowledge base is combined and the expert experience is utilized to guide model training and output, so that an expert experience-embedded power transmission line fault identification model is formed. And storing the model and applying the model to an online fault identification task. The unbalanced samples are expanded through an SMOTE method, so that the influence of the sample imbalance problem on the model performance is relieved; and an expert knowledge base is introduced into a model decision process, so that the accuracy and interpretability of the fault identification model are improved, and operation and maintenance personnel can easily understand an internal working mechanism of the model.

Description

Technical Field

[0001] The present invention relates to the field of power transmission line fault diagnosis and artificial intelligence deep learning technology, and in particular to a method for identifying the cause of a power transmission line fault based on experience guidance. Background Art

[0002] Transmission lines are key elements for the stable operation of power systems. As an important part of power systems, transmission lines are widely distributed in different regions, have complex operating environments, and are often affected by harsh environmental factors such as thunderstorms, strong winds, ice and snow, which can lead to various types of faults. Therefore, it is of great significance to quickly and accurately identify the causes of transmission line faults in order to guide operation and maintenance work and ensure the safe and reliable operation of power systems. With the development of machine learning and artificial intelligence technologies, artificial intelligence models such as support vector machines, deep belief networks, and convolutional neural networks have been introduced into the field of transmission line fault cause identification. They can combine the internal and external characteristics of faults based on in-depth analysis of various fault mechanisms to achieve high-accuracy fault identification. However, most of the existing fault identification schemes remain at the level of selecting the fault phase, and fail to deeply identify the real causes of transmission line faults (such as lightning strikes, wind deflection, wildfires, ice coverage, pollution flashover, foreign objects, etc.). On the other hand, for the solutions that have been carried out to identify the real causes of faults, most of them are direct applications of artificial intelligence models, which are prone to classification blind spots when faced with sample imbalance and increasingly complex system structures and operating conditions. Their universality needs to be further expanded, and the engineering experience of operation and maintenance personnel over the years has not been introduced into the model decision-making process, resulting in poor interpretability of the current solutions. Summary of the invention

[0003] The purpose of the present invention is to provide a method for identifying the cause of transmission line faults based on experience guidance. First, the unbalanced samples are expanded by the synthetic minority over-sampling technique (SMOTE) method; secondly, the implicit features in the fault transient waveform are fully extracted by the long short-term memory neural network (LSTM), and a nonlinear mapping relationship between the fault recording and the fault cause is preliminarily established; then, the engineering experience of the operation and maintenance personnel is sorted out, combined with the expert knowledge base and the expert experience is used to guide the model training and output, so as to form a transmission line fault identification model with embedded expert experience, thereby completing the fault identification task with high precision.

[0004] To achieve the above object, the present invention provides a method for identifying the cause of a power transmission line fault based on experience guidance, comprising the following steps:

[0005] S1. Expand the unbalanced samples by SMOTE method;

[0006] S2. The implicit features in the fault transient waveform are fully extracted through LSTM, and the nonlinear mapping relationship between the fault recording and the fault cause is preliminarily established;

[0007] S3. Organize the engineering experience of operation and maintenance personnel to form an expert knowledge base and use expert experience to guide model training to form a transmission line fault identification model with embedded expert experience.

[0008] Preferably, in step S1, assuming a certain unbalanced data set, for each data sample X in the minority class sample, search for its K nearest neighboring samples;

[0009] Assuming that the upsampling factor of the data set is n, n samples are randomly selected from the K nearest neighbor samples, denoted as the set Y = {y1, y2, ... y n};

[0010] Associate data samples X and Y, through X and y i The correlation formula between them is used for random interpolation to obtain the interpolation sample p i ;

[0011] For each data sample, n corresponding minority class samples are constructed; the interpolation formula is:

[0012] p i =X+r 0,1 ×(y i -X),i=1,2,…,n;

[0013] In the formula, X is the data sample in the minority class sample; r 0,1 is a random number in the interval (0,1); i Represents the i-th of the n nearest neighbor samples of data sample X.

[0014] Preferably, in step S2, the gating unit in the LSTM is composed of three parts, including an output gate, a forget gate and an output gate, and its expression is as follows:

[0015] f t =σ(W f ·[h t-1 ,x t ]+b f );

[0016] i t =σ(W i ·[h t-1 ,x t ]+b i );

[0017] O t =σ(W o ·[ht-1 ,x t ]+b o );

[0018] Where: f t Represents the state result of the forget gate, i t Represents the state result of the input gate, O t Represents the state result of the output gate, W f , W i , W o represents the corresponding connection weight, b f 、b i 、b o represents the corresponding bias term; x represents the input time series, x t Represents the input at the current moment, x t-1 represents the input at the previous moment, c t-1 Indicates the state of the gate control unit at the previous moment, h t Indicates the response output, c t represents the state of the gated unit; σ represents the activation function;

[0019] In LSTM, the forget gate determines c t-1 C t The input gate, forget gate and output gate are calculated, and the output gate and unit state jointly determine the unit output value, as shown below:

[0020]

[0021] h t =O t tanh(C t );

[0022] Where: b c is the bias term, W c is the state weight of the input unit, C t represents the unit state at time t, represents the cell state at time t, Represents the multiplication of corresponding elements of vectors of the same shape.

[0023] Preferably, in step S3, the power transmission line fault identification model based on expert experience is specifically as follows:

[0024] S31. Guide the fault identification model training and result output based on the engineering experience of the operation and maintenance personnel, so as to determine the cause of the fault. Based on the voltage and current recordings when the fault occurs, guide the model training and result output in combination with transition resistance, harmonic content, reclosing conditions, date information and meteorological description;

[0025] S32, the preliminary output result obtained by training the LSTM model in step S2, embedding the expert experience into the model, when completing the classification task, the LSTM model can be used for each sample x i (i is the sample number), and the preliminary probability p(y j |x i ), where y j is a certain fault cause category, with a total of J categories; satisfying:

[0026]

[0027] When expert experience is not embedded in the model, the model considers the maximum probability p(y j |x i ) belongs to the category label y j To output the result label, the cross entropy loss function is combined to realize back propagation and model iterative training.

[0028] Therefore, the present invention adopts a method for identifying the cause of a power transmission line fault based on experience guidance of the above structure, which has the following beneficial effects:

[0029] The present invention not only deeply identifies the real causes of transmission line failures (such as lightning strikes, wind deviation, wildfires, ice cover, pollution flashover, foreign objects, etc.), but also expands the unbalanced samples through the SMOTE method to alleviate the impact of sample imbalance on model performance; in addition, the engineering experience of operation and maintenance personnel over the years is organized into an expert knowledge base and introduced into the model decision-making process to improve the accuracy and interpretability of the fault identification model, making it easier for operation and maintenance personnel to understand the internal working mechanism of the model.

[0030] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 A detailed schematic diagram of an LSTM unit of a method for identifying the cause of a power transmission line fault based on experience guidance according to the present invention;

[0032] Figure 2 This is a LSTM timing diagram of a method for identifying the cause of a power transmission line fault based on experience guidance according to the present invention;

[0033] Figure 3 A schematic diagram of the technical idea of ​​embedding expert experience into an LSTM model of a power transmission line fault cause identification method based on experience guidance of the present invention;

[0034] Figure 4 It is a schematic diagram of the overall process of a method for identifying the cause of a power transmission line fault based on experience guidance of the present invention;

[0035] Figure 5 A schematic diagram of a confusion matrix of test results of a method for identifying the cause of a power transmission line fault based on experience guidance according to the present invention. DETAILED DESCRIPTION

[0036] The technical solution of the present invention is further described below through the accompanying drawings and embodiments.

[0037] Unless otherwise defined, the technical terms or scientific terms used in the present invention should be understood by people with ordinary skills in the field to which the present invention belongs. The words "first", "second" and similar words used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0038] Example

[0039] The present invention provides a method for identifying the cause of a transmission line fault based on experience guidance. S1. Unbalanced samples are expanded by using a SMOTE method.

[0040] Generally speaking, the causes of transmission line failures have a strong unbalanced attribute. Taking the transmission line failure cases that occurred in a certain place in recent years as an example, lightning strikes and foreign object failures are the main causes, accounting for 31.64% and 22.02% of the total failures respectively, flashover and wind deviation failures account for 15.50% and 13.54% of the total failures respectively, and other failures such as wildfires and ice cover account for about 5%. Transmission line failures caused by different reasons have different electrical characteristics and external properties. If the sample imbalance problem is not considered, it will increase the difficulty of intelligent model classification. To this end, the SMOTE method is used to expand the unbalanced samples, that is, random linear interpolation is performed between adjacent minority class samples to synthesize new minority class samples, thereby achieving the balance of the imbalanced data set. The specific features are as follows:

[0041] Assume an unbalanced data set. For each data sample X in the minority class, search for its K nearest neighbor samples (these K nearest neighbor samples all belong to the minority class). Assume that the upsampling factor of the data set is n, then randomly select n samples from the K nearest neighbor samples, denoted as the set Y = {y1, y2, …y n}. Associate data samples X and Y, through X and yi The correlation formula between them is used for random interpolation to obtain the interpolation sample p i . In this way, for each data sample, n corresponding minority class samples are constructed. The interpolation formula is:

[0042] p i =X+r 0,1 ×(y i -X),i=1,2,…,n;

[0043] In the formula, X is the data sample in the minority class sample; r 0,1 is a random number in the interval (0,1); i Represents the i-th of the n nearest neighbor samples of data sample X.

[0044] S2. Through LSTM, the implicit features in the fault transient waveform are fully extracted, and the nonlinear mapping relationship between the fault recording and the fault cause is preliminarily established.

[0045] LSTM is an improved recurrent neural network that can learn the implicit relationship between the time segments before and after the time series signal. At the same time, its internal memory unit and gate mechanism effectively overcome the gradient vanishing and explosion problems that may be caused by traditional recurrent neural network training, and realize the learning of long-distance dependent information. The details of its internal unit are as follows: Figure 1 , the expanded timing diagram is as follows Figure 2 .

[0046] In the figure, x represents the input time series, x t Represents the input at the current moment, x t-1 represents the input at the previous moment, c t-1 Indicates the state of the gate control unit at the previous moment, h t Indicates the response output, c t Represents the state of the gate control unit. Different from RNN, this gate control unit consists of three parts: the output gate, the forget gate, and the output gate. Its expression is as follows:

[0047] f t =σ(W f ·[h t-1 ,x t ]+b f );

[0048] i t =σ(W i ·[h t-1 ,x t ]+b i );

[0049] O t =σ(W o ·[h t-1,x t ]+b o );

[0050] Where: f t Represents the state result of the forget gate, i t Represents the state result of the input gate, O t Represents the state result of the output gate, W f , W i , W o represents the corresponding connection weight, b f 、b i 、b o represents the corresponding bias term; σ represents the activation function.

[0051] In LSTM, the forget gate determines c t-1 C t The input gate, forget gate and output gate are calculated, and the output gate and unit state jointly determine the unit output value, as shown below:

[0052]

[0053] h t =O t tanh(C t );

[0054] Where: b c is the bias term, W c is the state weight of the input unit, C t represents the unit state at time t, represents the cell state at time t, Represents the multiplication of corresponding elements of vectors of the same shape.

[0055] In comparison, the biggest advantage of the LSTM model is that it associates the previous and next input values ​​of its time series variables, so it can meet the requirements for processing time series variables. And by introducing the above-mentioned forgetting mechanism, it avoids the gradient disappearance or explosion that may be caused by RNN when processing long time series. The power system itself has obvious periodic characteristics, and its voltage / current waveform is a strong time series with strong correlation between the previous and next inputs. Therefore, the LSTM model is used in this scheme to construct the identification network.

[0056] S3. Organize the engineering experience of operation and maintenance personnel to form an expert knowledge base and use expert experience to guide model training to form a transmission line fault identification model with embedded expert experience.

[0057] S31, transmission line faults caused by different reasons have different electrical characteristics and external properties. Therefore, the fault identification model training and result output can be guided according to the engineering experience of the operation and maintenance personnel to determine the cause of the fault.

[0058] For example, lightning strike faults often occur in thunderstorms in summer and autumn. The transition resistance is mainly composed of pole tower resistance and arc resistance, which is often no more than 10Ω, and is a typical metallic grounding fault. The essence of foreign body fault is that objects with winding and covering properties are hung on the transmission line, causing the wire to short-circuit or form a discharge channel. Different foreign body materials have different conductive properties, which leads to complex transition resistance of foreign body faults, which can usually be divided into high-resistance foreign bodies and low-resistance foreign bodies. Pollution flashover faults are caused by the long-term accumulation of dirty things such as bird droppings on the surface of the insulator, which reduces the insulation resistance and generates local arcs in rainy, foggy and haze weather. Its transition resistance is greatly affected by the nonlinearity of arc resistance. Wind deviation faults often occur in winter and spring, and strong convection in summer is also prone to strong winds. When the wind force level reaches a certain level, the line deviates from the standard position and the air gap distance with objects such as poles and towers decreases, resulting in discharge. Since there are no other transition objects in the fault channel, it exhibits low resistance characteristics. Wildfire faults often occur in fire-prone areas such as dry forests in clear spring and autumn weather. The transition resistance is mainly composed of air thermal breakdown arc resistance and combustion resistance, and has obvious nonlinear high-resistance grounding characteristics. Icing faults often occur in low-temperature winter weather. As the air humidity increases, the surface of the transmission line is prone to condensation into rime or rime, which affects the insulation performance of the equipment, causing insulator breakdown and forming a discharge channel.

[0059] According to the above analysis, the model training and result output can be guided based on the voltage and current recordings when the fault occurs, combined with transition resistance, harmonic content, reclosing conditions, date information and meteorological description.

[0060] S32. For the preliminary output results obtained by training the LSTM model used in step 2, embed the expert experience into the model. Figure 3 shown.

[0061] When completing the classification task, the LSTM model can i (i is the sample number), and the preliminary probability p(y j |x i ), where y j is a certain fault cause category, with a total of J categories. Satisfying:

[0062]

[0063] Without embedding expert experience into the model, the model can consider the maximum probability p(y j |x i ) belongs to the category label y jThat is, the output result label is combined with the cross entropy loss function to realize back propagation and model iterative training. In this scheme, after the sample x is initially calculated, i The initial probability p(y j |x i ), the above probabilities are corrected by transition resistance, harmonic content, reclosing conditions, date information and meteorological description, for example:

[0064] 1) If the weather description contains the keyword "thunder", then for the sample x i The initial probability p(y j |x i ), the probability that the fault cause category is "lightning strike" is multiplied by a coefficient k1 greater than 1;

[0065] 2) If the fault transition resistance is large, then for the sample x i The initial probability p(y j |x i ), the probability that the fault cause category is “wildfire” is multiplied by a coefficient k2 greater than 1;

[0066] 3) If the fault occurs in the summer (or winter), then for the sample x i The initial probability p(y j |x i ), the probability that the fault cause category is "icing" is multiplied by a coefficient k3 that is less than (or greater than 1);

[0067] Through the above steps, the sample x i The initial probability p(y j |x i ) is converted into a modified probability Then all the probabilities of the J categories are normalized to satisfy:

[0068]

[0069] At this point, it can be considered that the maximum probability among J categories Category label y j That is, the output result.

[0070] In order to complete the task of high-precision transmission line fault cause identification, this scheme proposes a transmission line fault cause identification method based on experience guidance. The overall process is as follows: Figure 4 shown.

[0071] For each fault case, data preprocessing is performed. On the one hand, the time information, meteorological description and reclosing information of the fault location are obtained from the fault patrol report, and the corresponding feature encoding is performed; on the other hand, the three-phase voltage and current waveform data when the fault occurs are obtained from the fault recording file, and the two cycles from one cycle before the fault to the two cycles after the fault are intercepted, and the DC content and third harmonic content of the zero-sequence current are calculated based on this, and the fault transition resistance is calculated through the distance measurement information and line parameters. For the original time series input to LSTM, considering the differences in voltage level distribution and sampling frequency of the fault recording used, it is necessary to perform data homogeneity operation on it first, that is, first interpolate the time series data to unify it to the sampling frequency of 6.4kHz, and then normalize the interpolated time series data and scale it to the interval [-1,1] to avoid the amplitude and sampling frequency differences of the input signal affecting the model performance.

[0072] After preprocessing the fault cases, a deep learning model with better performance is trained offline using the processed fault sample set, and used for model testing applications. In the offline training stage, the data set is divided into a training set and a validation set according to a certain ratio. The training set is used to train the network parameters of the model, and the network hyperparameters are adjusted according to the classification effect of the model on the validation set to select the optimal performance model. During the training process, cross entropy is used as the loss function to quantify the difference between the model correction probability result and the true label; the Adam optimizer is used to calculate the minimum loss function, optimize the weights and biases of each layer of the model, and thus obtain the network model with the best performance. In the model testing application stage, after preprocessing the input data, the offline trained model is called for testing, and the final fault identification result can be output.

[0073] The data used in this paper are all from transmission line fault cases. According to the text records of the line inspection report, the fault causes are divided into 8 categories: lightning strike, foreign objects, pollution flashover, wind deviation, line break, wildfire, ice cover and others. The fault samples are divided into training set, validation set and test set according to the ratio of 4:2:1, and the fault samples are divided into training set, validation set and test set according to the ratio of 4:2:1. Figure 4 The process shown is input into the model. Considering that there is a certain sample imbalance in the test sample data set, the F1 score shown in the following formula is used to measure the classification accuracy of the fault identification model.

[0074]

[0075] Among them: True positive (T P ) represents the number of samples that are actually positive and predicted to be positive; false positive (F P ) represents the number of samples that are actually negative but predicted to be positive; false negative (F N ) represents the number of samples that are actually positive but predicted to be negative.

[0076] During the model training process, the number of iterations was set to 120, the learning rate α = 0.001, and the batch size β = 128. When the model training rounds reached 120, the classification accuracy of the fault identification model on the test samples reached 94.91%, and the F1 score was 92.78%. Its confusion matrix is ​​as follows Figure 5 As shown. Figure 5 It can be seen that the scheme proposed in this paper has a high recognition accuracy for fault types with a large number of samples such as lightning strikes, wind deviation, and foreign objects, and has a strong adaptability to a small number of samples such as icing, mountain fires, and pollution flashovers.

[0077] therefore,.

[0078] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A method for identifying the cause of a transmission line fault based on experience guidance, characterized in that: The following steps are involved: S1. Expand the unbalanced samples by SMOTE method; S2. The implicit features in the fault transient waveform are fully extracted through LSTM, and the nonlinear mapping relationship between the fault recording and the fault cause is preliminarily established; S3. Organize the engineering experience of operation and maintenance personnel to form an expert knowledge base and use expert experience to guide model training to form a transmission line fault identification model with embedded expert experience.

2. The method for identifying the cause of a power transmission line fault based on experience guidance according to claim 1, characterized in that: In step S1, assuming an unbalanced data set, for each data sample X in the minority class sample, search for its K nearest neighboring samples; Assuming that the upsampling factor of the data set is n, n samples are randomly selected from the K nearest neighbor samples, denoted as the set Y = {y1, y2, ... y n }; Associate data samples X and Y, through X and y i The correlation formula between them is used for random interpolation to obtain the interpolation sample p i ; For each data sample, n corresponding minority class samples are constructed; the interpolation formula is: p i =X+r 0,1 ×(y i -X),i=1,2,…,n; In the formula, X is the data sample in the minority class sample; r 0,1 is a random number in the interval (0,1); i Represents the i-th of the n nearest neighbor samples of data sample X.

3. The method for identifying the cause of a power transmission line fault based on experience guidance according to claim 1, characterized in that: In step S2, the gating unit in LSTM consists of three parts, including the output gate, the forget gate and the output gate, and its expression is as follows: f t =σ(W f ·[h t-1 ,x t ]+b f ); i t =σ(W i ·[h t-1 ,x t ]+b i ); The t =σ(W o ·[h t-1 ,x t ]+b o ); Where: f t Represents the state result of the forget gate, i t Represents the state result of the input gate, O t Represents the state result of the output gate, W f , W i , W o represents the corresponding connection weight, b f , b i , b o represents the corresponding bias term; x represents the input time series, x t Represents the input at the current moment, x t-1 represents the input at the previous moment, c t-1 Indicates the state of the gate control unit at the previous moment, h t Indicates the response output, c t represents the state of the gated unit; σ represents the activation function; In LSTM, the forget gate determines c t-1 C t The input gate, forget gate and output gate are calculated, and the output gate and unit state jointly determine the unit output value, as shown below: h t =O t ·tanh(C t ); Where: b c is the bias term, W c is the state weight of the input unit, C t represents the unit state at time t, represents the cell state at time t, Represents the multiplication of corresponding elements of vectors of the same shape.

4. The method for identifying the cause of a power transmission line fault based on experience guidance according to claim 3 is characterized in that: In step S3, the expert experience transmission line fault identification model is as follows: S31. Guide the fault identification model training and result output based on the engineering experience of the operation and maintenance personnel, so as to determine the cause of the fault. Based on the voltage and current recordings when the fault occurs, guide the model training and result output in combination with transition resistance, harmonic content, reclosing conditions, date information and meteorological description; S32, the preliminary output result obtained by training the LSTM model in step S2, embedding the expert experience into the model, when completing the classification task, the LSTM model can be used for each sample x i (i is the sample number), and the preliminary probability p(y j |x i ), where y j is a certain fault cause category, with a total of J categories; satisfying: When expert experience is not embedded in the model, the model considers the maximum probability p(y j |x i ) belongs to category label y j To output the result label, the cross entropy loss function is combined to realize back propagation and model iterative training.

Images