A method and system for detecting power grid line faults

Abstract

The present invention belongs to the technical field of power grid lines, and specifically relates to a power grid line fault detection method and system. The detection method comprises the following steps: obtaining training set current data and test set current data; introducing the Inceptionv3 structure, using the convolution kernel and BiLSTM in parallel to extract the power grid line fault features, inputting the feature fusion layer to form a joint feature, inputting the joint feature into the fully connected layer, performing time series prediction, and completing the construction of the power grid line fault prediction model; using the training set current data to train the prediction model; inputting the test set current data into the trained prediction model to realize the prediction of the feeder current at the next moment. The present invention accurately detects power grid line faults, ensures the safe operation of the power grid, and reduces the number of parameters and speeds up the operation by decomposing the convolution kernel in the model.

Description

Technical Field

[0001] The present invention belongs to the technical field of power grid lines, and in particular relates to a power grid line fault detection method and system. Background Art

[0002] Electricity is the cornerstone of a country's development. A stable supply of electricity can ensure the normal operation of industrial activities. During the operation of the power grid, it is often affected by many factors such as the human environment, the natural environment, and the social environment, leading to the occurrence of faults. According to historical statistics of power grid failures, distribution network failures account for 80% of all failures. As the final module of the power transmission system, the distribution network is directly connected to the user. Whether the distribution network can operate normally is related to the normal life and production of each user. At present, the main means of preventing and controlling feeder faults in the power grid still relies on manual detection. However, manual detection has the problems of high cost and lag, and cannot achieve real-time monitoring and detection.

[0003] In recent years, deep learning has been rapidly applied in various fields due to its powerful data processing and self-learning capabilities. In the field of power grid line fault detection, deep learning provides an intelligent foundation for fault detection. By establishing a deep learning model, the inherent implicit features in the data are mined to achieve early identification of power grid line faults, effectively detect distribution network line faults, timely reduce the occurrence of faults, improve the reliability of distribution network power distribution, reasonably optimize power grid equipment, save manpower, and ensure the rigid needs of users in production or life. Zhang Pengfei et al. disclosed a distribution network fault number prediction based on meteorological factors and time series analysis. Using the distribution network fault repair data of a city in East China, a short-term prediction study on the number of distribution network faults was carried out. A seasonal determination method based on temperature was proposed. Multivariate regression and time series analysis were used to construct a seasonal meteorological impact fault prediction model, determine the quantitative relationship between meteorological factors such as temperature and weather and the fault volume, and construct an autoregressive moving average time series prediction model for the remaining fault volume after eliminating the influence of meteorological factors to capture the time series change trend of the fault volume. Through the comprehensive application of the above models, the short-term prediction of the number of distribution network faults was achieved. However, this method can only detect the number of faults at a certain time to formulate emergency repair plans, but it cannot directly locate the faulty line or branch line, and cannot quickly troubleshoot the faulty line. Summary of the invention

[0004] To overcome the above-mentioned defects, the purpose of the present invention is to provide a power grid line fault detection method and system, which can directly locate the faulty line with high positioning accuracy, directly detect the health status of each power grid line, and accurately diagnose power grid line faults to ensure the normal operation of the power grid.

[0005] To achieve the above object, the present invention adopts the following technical solution:

[0006] A method for detecting a power grid line fault comprises the following steps:

[0007] S1. Obtain training set current data and test set current data;

[0008] S2. Introduce the Inceptionv3 structure, use the convolution kernel and BiLSTM in parallel to extract the power line fault features, input the feature fusion layer to form joint features, input the joint features to the fully connected layer, predict the time series, and complete the construction of the power line fault prediction model ImTreNet;

[0009] S3. Input the training set current data as a training set into the ImTreNet model built in step S2 to train the model;

[0010] S4. Input the test set current data into the prediction model trained in step S3 to predict the feeder current at the next moment.

[0011] Preferably, the step S1 of acquiring the training set current data and the test set current data includes:

[0012] Determine the ID of the feeder to be tested and use its current data as the target parameter, select other feeder currents with a correlation coefficient greater than 0.5 with the feeder under the same substation as input parameters; standardize the current data of each feeder to obtain training set current data and test set current data.

[0013] Preferably, the correlation coefficient is the Pearson correlation coefficient, and the Pearson correlation coefficient calculation formula is as follows:

[0014]

[0015] In the formula, r is the Pearson correlation coefficient, n is the number of samples, and x is i and i is the i-th sampling point, and is the sample mean.

[0016] Preferably, the standardization processing method is:

[0017]

[0018] Where x' represents the standardized current data, x represents the current data, μ is the mean, and σ is the variance.

[0019] Preferably, the convolution kernel described in step S2 is a one-dimensional convolution kernel with sizes of (1, 1), (3, 3), (1, 3), and (3, 1).

[0020] Preferably, the information propagation of the BiLSTM described in step S2 is divided into two parts, forward and backward, and the output at each moment includes both the forward output and the backward output of the neuron, and the calculation formula is as follows:

[0021] h t =f(w1x t +w2h t-1 +b1)

[0022] h′ t =f(w3x t +w5h t+1 +b2)

[0023] o′ t =g(w4h t +b3)

[0024] o″ t =g(w6h′ t +b4)

[0025] O t =(o′ t +o″ t ) / 2

[0026] In the formula, f and g represent activation functions, w i (i=1…6) represents the weight from one network layer to another network layer; {…h t-1 ,h t ,h t+1 ...} represents the forward propagation layer in the BiLSTM neural network, {...x t-1 ,x t ,x t+1 ...} represents the input of the forward propagation layer; {...h t+1 ′,h t ′,h t-1 ′…} represents the back-propagation layer in the BiLSTM neural network, {…x t+1 ,x t ,x t-1 …} represents the input of the back propagation layer; b1, b2, b3, b4 represent the bias coefficients of the hidden units, o′ t , o′ t The result of two LSTM units processing the input feature vectors at the corresponding time is that the two feature vectors at the corresponding time are added and the average value is taken as the output time series feature vector O t .

[0027] Preferably, the step S4 of inputting the test set current data into the trained prediction model to achieve the prediction of the feeder current at the next moment includes:

[0028] The root mean square error (RMSE) between the actual value of the feeder current and the model output value is used as a detection index of the power grid line fault to detect the power grid fault;

[0029] The calculation formula of the RMSE value is:

[0030]

[0031] In the formula, R is the root mean square error RMSE value, r i is the actual value, is the model output value, and n is the number of samples.

[0032] Preferably, the detection of the power grid fault further comprises: comparing the RMSE value with a state detection threshold to detect the power grid fault;

[0033] The calculation formula of the state detection threshold is:

[0034]

[0035] Where L(t) is the state detection threshold, μ R is the mean of R, σ R is the standard deviation of R, n S is the sampling length, z is the location coefficient, and β is the weight of historical R on the current statistic.

[0036] A power grid line fault detection system, which implements the above detection method, comprises:

[0037] A data acquisition module, which acquires current data of a training set and current data of a test set;

[0038] The model building module introduces the Inceptionv3 structure, uses the convolution kernel and BiLSTM in parallel to extract the power line fault features, inputs the feature fusion layer to form joint features, inputs the joint features to the fully connected layer, predicts the time series, and completes the construction of the power line fault prediction model ImTreNet.

[0039] The model training module inputs the training set current data into the built ImTreNet model as a training set to train the model;

[0040] The fault detection module inputs the test set current data into the trained prediction model to predict the feeder current at the next moment.

[0041] Preferably, the data acquisition module is specifically used for:

[0042] Determine the ID of the feeder to be tested and use its current data as the target parameter, select other feeder currents with a correlation coefficient greater than 0.5 with the feeder under the same substation as input parameters; standardize the current data of each feeder to obtain training set current data and test set current data.

[0043] Preferably, the correlation coefficient is the Pearson correlation coefficient, and the Pearson correlation coefficient calculation formula is as follows:

[0044]

[0045] In the formula, r is the Pearson correlation coefficient, n is the number of samples, and x is i and i is the i-th sampling point, and is the sample mean.

[0046] Preferably, the standardization processing method is:

[0047]

[0048] Where x' represents the standardized current data, x represents the current data, μ is the mean, and σ is the variance.

[0049] Preferably, the convolution kernel is a one-dimensional convolution kernel with sizes of (1, 1), (3, 3), (1, 3), and (3, 1).

[0050] Preferably, the information propagation of the BiLSTM is divided into two parts, forward and backward. The output at each moment includes both the forward output and the backward output of the neuron. The calculation formula is as follows:

[0051] h t =f(w1x t +w2h t-1 +b1)

[0052] h′ t =f(w3x t +w5h t+1 +b2)

[0053] o′ t =g(w4h t +b3)

[0054] o″ t =g(w6h′ t +b4)

[0055] O t =(o′ t +o″ t ) / 2

[0056] In the formula, f and g represent activation functions, w i (i=1…6) represents the weight from one network layer to another network layer; {…h t-1 ,h t ,h t+1 ...} represents the forward propagation layer in the BiLSTM neural network, {...x t-1 ,x t ,x t+1 ...} represents the input of the forward propagation layer; {...h t+1 ′,h t ′,h t-1 ′…} represents the back-propagation layer in the BiLSTM neural network, {…x t+1 ,x t ,x t-1 …} represents the input of the back propagation layer; b1, b2, b3, b4 represent the bias coefficients of the hidden units, o′ t ,o″ t The result of two LSTM units processing the input feature vectors at the corresponding time is that the two feature vectors at the corresponding time are added and the average value is taken as the output time series feature vector O t .

[0057] Preferably, the fault detection module is specifically used for:

[0058] The root mean square error (RMSE) between the actual value of the feeder current and the model output value is used as a detection index of the power grid line fault to detect the power grid fault;

[0059] The calculation formula of the RMSE value is:

[0060]

[0061] In the formula, R is the root mean square error RMSE value, r i is the actual value, is the model output value, and n is the number of samples.

[0062] Preferably, the detection of the power grid fault further comprises: comparing the RMSE value with a state detection threshold to detect the power grid fault;

[0063] The calculation formula of the state detection threshold is:

[0064]

[0065] Where L(t) is the state detection threshold, μ R is the mean of R, σ R is the standard deviation of R, n S is the sampling length, z is the location coefficient, and β is the weight of historical R on the current statistic.

[0066] The positive beneficial effects of the present invention are:

[0067] 1. The present invention discloses a power grid line fault detection method based on a deep learning model, introduces the inceptionv3 structure, and reduces the number of parameters and speeds up the operation by decomposing the convolution kernel in the model.

[0068] 2. The present invention uses convolution kernels of different sizes to extract local features. Each layer of convolution will learn the entire data set, improve the local feature extraction ability of the model, and ensure that accurate local features are obtained. At the same time, considering the long-term dependence in the historical trend sequence, the problems of "gradient disappearance" and "long-term dependence" are also avoided. The ImTreNet model uses a bidirectional long short-term memory network BiLSTM to capture this dependence. BiLSTM extracts feature information from both positive and negative directions by constructing two LSTMs, fully learning the temporal characteristics of the data, and combining the output of the convolution kernel and BiLSTM through the feature fusion layer to form a joint feature. The joint feature is input into the fully connected layer for time series prediction, thereby accurately detecting power grid line faults and ensuring the safe operation of the power grid.

[0069] 3. The present invention first trains the model based on the current data of the normal period, and then inputs the current data collected by the power grid line into the model. If an early fault occurs, the root mean square error between the model prediction value and the true value will increase significantly. The threshold is set by the root mean square error value calculated by the predicted value and the true value under the normal operation of the power grid line. When the root mean square error between the predicted value and the true value exceeds the threshold, an alarm will be generated on the faulty line. The detection method of the present invention directly locates the faulty line, has high positioning accuracy, and directly detects the health status of each power grid line, making it more convenient for line operation and inspection personnel to conduct fault inspection and troubleshooting, greatly reducing the inspection workload of the operation and inspection personnel. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 It is a flow chart of the power grid line fault detection method of the present invention;

[0071] Figure 2 This is the structure diagram of the ImTreNet model;

[0072] Figure 3 Schematic diagram of BiLSTM neural network;

[0073] Figure 4 is the time domain diagram of feeder current;

[0074] Figure 5 This is a diagram for identifying faults in power grid lines. DETAILED DESCRIPTION

[0075] The present invention is further described below in conjunction with some specific embodiments.

[0076] Example 1

[0077] See also Figure 1 , a power grid line fault detection method, comprising the following steps:

[0078] S1. Get the training set current data and the test set current data:

[0079] 1) Obtain the current data of different feeders under the same substation, determine the ID of the feeder to be tested and use its current data as the target parameter. In order to further compress the data volume and make the information of the input data more comprehensive, correlation analysis is used to select more valuable variables from each feeder. The purpose of this operation is to calculate the correlation between each variable through the Pearson correlation coefficient, and select the healthy feeder current data with a strong correlation with the target parameter as the final input of the model. The present invention selects other feeder currents under the same substation with a correlation coefficient greater than 0.5 with the above-mentioned faulty feeder as input parameters. The correlation coefficient adopts the Pearson correlation coefficient. The calculation formula of the Pearson correlation coefficient is shown as follows:

[0080]

[0081] In the formula, r is the Pearson correlation coefficient, n is the number of samples, and x is i and i is the i-th sampling point, and is the sample mean.

[0082] 2) Standardizing the current data of each feeder and dividing the standardized data into training set current data and test set current data;

[0083] The specific method for standardizing current data is as follows:

[0084]

[0085] Where x' represents the standardized current data, x represents the current data, μ is the mean, and σ is the variance.

[0086] S2. Construction of ImTreNet model:

[0087] Local feature extraction is performed using one-dimensional convolution kernels of different sizes. The convolution kernel sizes are (1, 1), (3, 3), (1, 3), and (3, 1). The bidirectional long short-term memory network BiLSTM is used for global feature extraction. The Inceptionv3 structure is introduced, and the convolution kernel and BiLSTM are used in parallel to fully realize the extraction of power line fault features. The feature fusion layer is used to form joint features, which are input into the fully connected layer for time series prediction. The construction of the power line fault prediction model ImTreNet is completed. See Figure 2 ;

[0088] The convolution kernel and BiLSTM network parameter settings are shown in Table 1.

[0089] Table 1 ImTreNet network parameter settings

[0090]

[0091] S3. Training of ImTreNet model

[0092] The current data of the feeder in a healthy state is input into the built ImTreNet model as a training set to train the model;

[0093] In the training phase, the sequence T is input into the BiLSTM layer in ImTreNet, see Figure 3 , BiLSTM information propagation is divided into two parts: forward and backward. The output at each moment contains both the forward output and backward output of the neuron. The BiLSTM calculation formula is as follows:

[0094] h t =f(w1x t +w2h t-1 +b1)

[0095] h t ′t=f(w3x t +w5h t+1 +b2)

[0096] o′ t =g(w4h t +b3)

[0097] o″ t =g(w6h t ′+b4)

[0098] O t =(o′ t +o″ t ) / 2

[0099] In the formula, f and g represent activation functions, wi (i=1…6) represents the weight from one network layer to another network layer; {…h t-1 ,h t ,h t+1 ...} represents the forward propagation layer of BiLSTM, {...x t-1 ,x t ,x t+1 ...} represents the input of the forward propagation layer; {...h t+1 ′,h t ′,h t-1 …} represents the back-propagation layer in the BiLSTM neural network, {…x t+1 ,x t ,x t-1 ...} represents the input of the back-propagation layer; b1, b2, b3, b4 represent the bias coefficients of the hidden units.

[0100] '"

[0101] Number, o t , o t The result of two LSTM units processing the input feature vectors at the corresponding time is that the two feature vectors at the corresponding time are added and the average value is taken as the output time series feature vector O t .

[0102] See also Figure 2 When the sequence T is input into BiLSTM, the BiLSTM layer is used to extract the time features contained in the signal, and the sequence T is input into three branches containing convolution operations respectively. The first branch performs a 1×1 convolution operation on the input. The 1×1 convolution can organize information across channels and improve the expression ability of the network. The second branch first uses a 1×1 convolution and then a 3×3 convolution. The present invention uses the inceptionv3 structure to further decompose the 3×3 convolution in the second branch into 1×3 and 3×1 convolutions. The improved convolution kernel speeds up the operation when processing large feature data. The third branch first uses 1×1 convolution and then 5×5 convolution. The inceptionv3 structure is used to replace the 5×5 convolution kernel with two 3×3 convolution kernels, and one of the 3×3 convolution kernels is further split into 1×3 and 3×1 convolution kernels, which not only reduces the number of parameters but also improves the calculation speed. Each branch will learn the entire data set to ensure accurate local features, and then the feature fusion layer combines the output of BiLSTM and the convolution kernel to form a joint feature, which is input into the fully connected layer for time series prediction to complete the model training.

[0103] S4. Grid line fault detection:

[0104] The test set data is input into the trained model. The residual between the actual value of the feeder current and the model output value always exists. The residual value reflects the health level of the line. In order to more accurately reflect the change of the residual, the RMSE value is calculated in days. By observing the change trend of the RMSE value, the fault detection of the power grid is realized. The RMSE value calculation formula is:

[0105]

[0106] In the formula, R is the root mean square error RMSE value, r i is the actual value, is the model output value, and n is the number of samples.

[0107] When the power grid line is operating normally, the input data will match the model well, the output residual will be small enough, and the RMSE value will also be small enough; on the contrary, when the line is in an unhealthy or faulty state, the data will not adapt to the model parameters, and the RMSE value will be much larger than the training error. Therefore, the RMSE value is used as a detection indicator.

[0108] Since the model is trained by the data of the feeder in normal state, it can reflect the characteristics of the feeder during normal operation. In order to further improve the expression ability of the model, it is necessary to set the state detection threshold of the model. In order to define the state detection threshold of the fault, EWMA is used to calculate.

[0109] Exponentially weighted moving average (EWMA) is a statistical method that can smooth data sets. As time goes by, its computational complexity will decrease. In addition, EWMA is also sensitive to small changes. The RMSE value of the deep learning model is processed using the EWMA method, and the RMSE value of the healthy period data is used to calculate the warning threshold and alarm threshold. The calculation method of EWMA is:

[0110] v t =βR t +(1-β)v t-1

[0111] In the formula, v t is the statistic of the EWMA control line, R t is the RMSE value at time t, β is the weight of historical R to the current statistic, and its value is 0.2 during the analysis.

[0112] v t The mean and variance of can be calculated by the following formula:

[0113]

[0114] In the formula, μ Ris the mean of R, σR is the standard deviation of R, nS is the sampling length, and the sampling length selected in this analysis process is 48.

[0115] The calculation formula for the status detection threshold is as follows:

[0116]

[0117] μ R and σ R Substituting into L(t) we get the calculation formula for the state detection threshold:

[0118]

[0119] Where L(t) is the state detection threshold of the t-th time node, μ R is the mean of R, σR is the standard deviation of R, nS is the sampling length, β is the weight of historical R to the current statistic, and its value is 0.2; z is the location coefficient. In order to prevent false alarms, z values ​​are taken as 2 and 28 respectively. When z is 2, the warning threshold of EWMA is calculated; when z is 28, the alarm threshold of EWMA is calculated. When RMSE exceeds the warning threshold, it indicates that a fault may have occurred. When RMSE exceeds the alarm threshold, it indicates that an early fault has occurred.

[0120] The effectiveness of the present invention is verified by analyzing the current data of each feeder of a substation for a whole year.

[0121] The current data of each feeder of a substation in 2021 are obtained. The feeder IDs are 17DKX-329377, 17DKX-327489, 17DKX-479793, 17DKX-330513, 17DKX-773761, 17DKX-908801, 17DKX-1563265, 17DKX-648657, 17DKX-909137, and 17DKX-788801. Among them, the ID of the faulty feeder is 17DKX-329377. This feeder is selected as the target parameter. The Pearson correlation coefficients of 17DKX-329377 and other feeders are shown in Table 2. It can be found from Table 2 that the correlation coefficients of other feeders with the faulty feeder are all greater than 0.5.

[0122] Table 2 Correlation coefficients between feeders

[0123] Feeder ID 17DKX-329377 17DKX-327489 17DKX-479793 17DKX-330513 17DKX-773761 Correlation coefficient 1 0.680364356 0.52682804 0.645846135 0.695869624 Feeder ID 17DKX-908801 17DKX-1563265 17DKX-648657 17DKX-909137 17DKX-788801 Correlation coefficient 0.58138926 0.646336981 0.621193265 0.657489016 0.599987803

[0124] Through the above steps, the data is preprocessed and the model is built. MSE, RMSE, R 2The three evaluation indicators are used to evaluate the prediction performance of the model, and the indicators are shown in Table 3. The ImTreNet model shows good performance in all indicators. The MSE and RMSE error indicators are both small, and R 2 It is close to 1, which means that the predicted value of the power grid in a healthy state is close to the true value, and the model has a good fitting degree, which further shows that the model can achieve accurate prediction when the power grid is in normal operation.

[0125] Table 3 ImTreNet model evaluation index table

[0126] Evaluation indicators MSE RMSE <![CDATA[R 2 ]]> ImTreNet 0.155 0.393 0.934

[0127] Figure 4 The time domain diagram of the feeder current generating the fault is given by Figure 4 It can be seen that the current tended to be stable before 160 days, and then began to increase continuously after 160 days. A fault occurred at 220 days and the power was shut down for maintenance. After the maintenance, the current returned to normal. Figure 5 This is the fault identification diagram of the power grid line. Figure 5 It can be seen that the RMSE value reached the warning threshold at 160 days, reached the alarm threshold at 170 days, and returned to the normal level after 220 days of maintenance, indicating that the model can effectively detect power grid line faults.

[0128] Example 2

[0129] A power grid line fault detection system, used to implement the detection method described in Example 1, comprising:

[0130] A data acquisition module, which acquires current data of a training set and current data of a test set;

[0131] The model building module introduces the Inceptionv3 structure, uses the convolution kernel and BiLSTM in parallel to extract the power line fault features, inputs the feature fusion layer to form joint features, inputs the joint features to the fully connected layer, predicts the time series, and completes the construction of the power line fault prediction model ImTreNet.

[0132] The model training module inputs the training set current data into the built ImTreNet model as a training set to train the model;

[0133] The fault detection module inputs the test set current data into the trained prediction model to predict the feeder current at the next moment.

[0134] Furthermore, the data acquisition module is specifically used for:

[0135] Determine the ID of the feeder to be tested and use its current data as the target parameter, select other feeder currents with a correlation coefficient greater than 0.5 with the feeder under the same substation as input parameters; standardize the current data of each feeder to obtain training set current data and test set current data.

[0136] Furthermore, the correlation coefficient is the Pearson correlation coefficient, and the calculation formula of the Pearson correlation coefficient is as follows:

[0137]

[0138] In the formula, r is the Pearson correlation coefficient, n is the number of samples, and x is i and i is the i-th sampling point, and is the sample mean.

[0139] Furthermore, the standardization method is:

[0140]

[0141] Where x' represents the standardized current data, x represents the current data, μ is the mean, and σ is the variance.

[0142] Furthermore, the convolution kernel is a one-dimensional convolution kernel with sizes of (1, 1), (3, 3), (1, 3), and (3, 1) to perform multi-scale local abstract feature extraction.

[0143] Furthermore, the information propagation of BiLSTM is divided into two parts: forward and backward. The output at each moment contains both the forward output and backward output of the neuron. The calculation formula is as follows:

[0144] h t =f(w1x t +w2h t-1 +b1)

[0145] h′ t =f(w3x t +w5h t+1 +b2)

[0146] o′ t =g(w4h t +b3)

[0147] o″ t =g(w6h′ t +b4)

[0148] O t =(o′ t +o″ t ) / 2

[0149] In the formula, f and g represent activation functions, w i (i=1…6) represents the weight from one network layer to another network layer; {…h t-1 ,h t ,h t+1 ...} represents the forward propagation layer in the BiLSTM neural network, {...x t-1 ,x t ,x t+1 ...} represents the input of the forward propagation layer; {...h t+1 ′,h t ′,h t-1 ′…} represents the back-propagation layer in the BiLSTM neural network, {…x t+1 ,x t ,x t-1 …} represents the input of the back propagation layer; b1, b2, b3, b4 represent the bias coefficients of the hidden units, o′ t ,o″ t The result of two LSTM units processing the input feature vectors at the corresponding time is that the two feature vectors at the corresponding time are added and the average value is taken as the output time series feature vector O t .

[0150] Furthermore, the fault detection module is specifically used for:

[0151] The root mean square error (RMSE) between the actual value of the feeder current and the model output value is used as a detection index of the power grid line fault to detect the power grid fault;

[0152] The calculation formula of RMSE value is:

[0153]

[0154] In the formula, R is the root mean square error RMSE value, r i is the actual value, is the model output value, and n is the number of samples.

[0155] Furthermore, the detection of the power grid fault further includes: comparing the RMSE value with a state detection threshold to detect the power grid fault;

[0156] The calculation formula for the state detection threshold is:

[0157]

[0158] Where L(t) is the state detection threshold, μ R is the mean of R, σ R is the standard deviation of R, n Sis the sampling length, z is the location coefficient, and β is the weight of historical R on the current statistic.

[0159] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes. The schemes in the embodiments of the present invention may be implemented in various computer languages, for example, object-oriented programming language Java and literal scripting language JavaScript, etc.

[0160] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems) and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the process in the flowchart. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0161] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0162] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0163] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0164] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A method for detecting a power grid line fault, characterized in that: The following steps are involved: S1. Obtain training set current data and test set current data; S2. Introduce the Inceptionv3 structure, use the convolution kernel and BiLSTM in parallel to extract the power line fault features, input the feature fusion layer to form joint features, input the joint features to the fully connected layer, predict the time series, and complete the construction of the power line fault prediction model ImTreNet; S3. Input the training set current data as a training set into the ImTreNet model built in step S2 to train the model; S4. Input the test set current data into the prediction model trained in step S3 to predict the feeder current at the next moment.

2. The power grid line fault detection method according to claim 1, characterized in that: The step S1 of obtaining the training set current data and the test set current data includes: Determine the ID of the feeder to be tested and use its current data as the target parameter, select other feeder currents with a correlation coefficient greater than 0.5 with the feeder under the same substation as input parameters; standardize the current data of each feeder to obtain training set current data and test set current data.

3. The power grid line fault detection method according to claim 2, characterized in that: The correlation coefficient is the Pearson correlation coefficient, and the Pearson correlation coefficient calculation formula is as follows: In the formula, r is the Pearson correlation coefficient, n is the number of samples, and x is i and i is the i-th sampling point, and is the sample mean.

4. The power grid line fault detection method according to claim 2, characterized in that: The standardized processing method is: Where x' represents the standardized current data, x represents the current data, μ is the mean, and σ is the variance.

5. The power grid line fault detection method according to claim 1, characterized in that: The convolution kernel described in step S2 is a one-dimensional convolution kernel with sizes of (1, 1), (3, 3), (1, 3), and (3, 1).

6. The power grid line fault detection method according to claim 1, characterized in that: The information propagation of the BiLSTM described in step S2 is divided into two parts: forward and backward. The output at each moment includes both the forward output and the backward output of the LSTM unit. The calculation formula is as follows: h t =f(w1x t +w2h t-1 +b1) h′ t =f(w3x t +w5h t+1 +b2) the' t =g(w4h t +b3) o″ t =g(w6h′ t +b4) The t =(o′ t +o″ t ) / 2 In the formula, f and g represent activation functions, w i represents the weight from one network layer to another network layer, i = 1...6; {...h t-1 ,h t ,h t+1 ...} represents the forward propagation layer in the BiLSTM neural network, {...x t-1 , x t , x t+1 ...} represents the input of the forward propagation layer; {...h t+1 ′,h t ′,h t-1 ′…} represents the back-propagation layer in the BiLSTM neural network, {…x t+1 , x t , x t-1 …} represents the input of the back propagation layer; b1, b2, b3, b4 represent the bias coefficients of the hidden units, o′ t ,o″ t The result of two LSTM units processing the input feature vectors at the corresponding time is that the two feature vectors at the corresponding time are added and the average value is taken as the output time series feature vector O t .

7. The power grid line fault detection method according to claim 1, characterized in that: Step S4 of inputting the test set current data into the trained prediction model to achieve the prediction of the feeder current at the next moment includes: The root mean square error (RMSE) between the actual value of the feeder current and the model output value is used as a detection index of the power grid line fault to detect the power grid fault; The calculation formula of the RMSE value is: In the formula, R is the root mean square error RMSE value, r i is the actual value, is the model output value, and n is the number of samples.

8. The power grid line fault detection method according to claim 7, characterized in that: The detection of the power grid fault further includes: comparing the RMSE value with a state detection threshold to detect the power grid fault; The calculation formula of the state detection threshold is: Where L(t) is the state detection threshold, μ R is the mean of R, σ R is the standard deviation of R, n S is the sampling length, z is the location coefficient, and β is the weight of historical R on the current statistic.

9. A power grid line fault detection system, characterized in that: include: A data acquisition module, which acquires current data of a training set and current data of a test set; The model building module introduces the Inceptionv3 structure, uses the convolution kernel and BiLSTM in parallel to extract the power line fault features, inputs the feature fusion layer to form joint features, inputs the joint features to the fully connected layer, predicts the time series, and completes the construction of the power line fault prediction model ImTreNet. The model training module inputs the training set current data into the built ImTreNet model as a training set to train the model; The fault detection module inputs the test set current data into the trained prediction model to predict the feeder current at the next moment.

10. The power grid line fault detection system according to claim 9, characterized in that: The data acquisition module is specifically used to: determine the ID of the feeder to be tested and use its current data as the target parameter, and select other feeder currents under the same substation with a correlation coefficient greater than 0.5 of the feeder as input parameters; The current data of each feeder is standardized to obtain the training set current data and the test set current data.

11. The power grid line fault detection system according to claim 10, characterized in that: The correlation coefficient is the Pearson correlation coefficient, and the Pearson correlation coefficient calculation formula is as follows: In the formula, r is the Pearson correlation coefficient, n is the number of samples, and x is i and i is the i-th sampling point, and is the sample mean.

12. The power grid line fault detection system according to claim 10, characterized in that: The standardized processing method is: Where x' represents the standardized current data, x represents the current data, μ is the mean, and σ is the variance.

13. The power grid line fault detection system according to claim 9, characterized in that: The convolution kernel is a one-dimensional convolution kernel with sizes of (1, 1), (3, 3), (1, 3), and (3, 1).

14. The power grid line fault detection system according to claim 9, characterized in that: The information propagation of the BiLSTM is divided into two parts: forward and backward. The output at each moment contains both the forward output and backward output of the LSTM unit. The calculation formula is as follows: h t =f(w1x t +w2h t-1 +b1) h t ′=f(w3x t +w5h t+1 +b2) the' t =g(w4h t +b3) o′ t ′=g(w6h t ′+b4) The t =(o′ t +o′ t ′) / 2 In the formula, f and g represent activation functions, w i represents the weight from one network layer to another network layer, i = 1...6; {...h t-1 ,h t ,h t+1 ...} represents the forward propagation layer in the BiLSTM neural network, {...x t-1 ,x t ,x t+1 ...} represents the input of the forward propagation layer; {...h t+1 ′,h t ′,h t-1 ′…} represents the back-propagation layer in the BiLSTM neural network, {…x t+1 ,x t ,x t-1 ...} represents the input of the back propagation layer; b1, b2, b3, b4 represent the bias coefficients of the hidden units ″′ Number, o t , o t The result of two LSTM units processing the input feature vectors at the corresponding time is that the two feature vectors at the corresponding time are added and the average value is taken as the output time series feature vector O t .

15. The power grid line fault detection system according to claim 9, characterized in that: The fault detection module is specifically used to: use the root mean square error RMSE value between the actual value of the feeder current and the model output value as a detection index of the power grid line fault to detect the power grid fault; The calculation formula of the RMSE value is: In the formula, R is the root mean square error RMSE value, r i is the actual value, is the model output value, and n is the number of samples.

16. The power grid line fault detection system according to claim 15, characterized in that: The detection of the power grid fault further includes: comparing the RMSE value with a state detection threshold to detect the power grid fault; The calculation formula of the state detection threshold is: Where L(t) is the state detection threshold, μ R is the mean of R, σ R is the standard deviation of R, n S is the sampling length, z is the location coefficient, and β is the weight of historical R on the current statistic.

Images