Intelligent power distribution monitoring and early warning method

Abstract

The present invention belongs to the field of power distribution warning technology, and in particular, relates to a method for intelligent power distribution monitoring and warning. First, the system obtains the temperature, current and power parameters of the distribution line with time series characteristics through temperature sensors, current sensors and power sensors, removes outliers, noise and unreliable data through Savitzky-Go-Lay filters, eliminates conflicts and environmental interference between time series data, and then builds an improved long and short-term memory network fault diagnosis model, classifies the data according to power parameters, and then integrates the information from the sensor through the long and short-term memory network for data fusion, introduces a memory unit structure, and finally uses the trained model to diagnose the distribution line fault and transmits the results to the controller, which controls the alarm and communication system. The results of the monitoring and early warning method proposed in this patent are more accurate, and the comprehensive diagnosis effect is the best.

Description

Technical Field

[0001] The present invention belongs to the technical field of power distribution early warning, and in particular relates to a method for intelligent power distribution monitoring and early warning. Background Art

[0002] The intelligent power distribution system can realize real-time monitoring of the operating status of the power grid, improve power supply reliability and quality, and is an important part of the power system. Once an interruption occurs, it will seriously affect people’s natural life and production. Therefore, timely and accurate monitoring and early warning of power distribution faults are of great significance to ensure the reliability and safety of the distribution network.

[0003] At present, the monitoring and early warning method of the intelligent distribution system is mainly the neural network method. However, the diagnosis process of a single neural network method may take a certain amount of time to analyze data and generate results, which may lead to inaccurate diagnosis and timely fault handling. In addition, some normal situations may be misjudged or misreported as faults based on a single parameter, which may lead to unnecessary repairs and interventions, increasing the maintenance cost of the system. Summary of the invention

[0004] In view of the technical problems existing in power distribution early warning, the present invention proposes an intelligent power distribution monitoring and early warning method with reasonable design, simple method, strong theoretical basis, fast response, strong anti-interference ability, more accurate diagnosis result and best comprehensive diagnostic effect.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is: an intelligent power distribution monitoring and early warning method, comprising the following steps:

[0006] S1. The system conducts self-inspection and obtains the temperature, current and power parameters with time sequence characteristics of all distribution cables;

[0007] S2, remove outliers, noise and unreliable data through Savitzky-Go-Lay filter, and eliminate conflicts and environmental interference between time series data;

[0008] S3. Build a fault diagnosis model based on the long short-term memory network. Considering that different power consumption modes use different line parameters, introduce an improved K-means algorithm before the long short-term memory network to divide power distribution into four categories: industrial power, agricultural power, commercial power, and municipal life power. The improved K-means algorithm adds Fréchet distance, Jaccard similarity coefficient, and cosine similarity to the Euclidean distance to construct a comprehensive distance. The formula for calculating the shortest distance between each cluster object and the cluster center is:

[0009] D(c k ,x i )=min{α1ED(c k ,xi )+α2FD(c k ,x i )+α3JD(c k ,x i )+α4TD(c k ,x i )}

[0010] Among them, c k represents the kth cluster center, x i Indicates that the clustering object is a collection of temperature, current and power parameters, ED(c k ,x i ) means c k and x i The Euclidean distance, FD(c k ,x i ) means c k and x i The Fréchet distance, JD(c k ,x i ) means c k and x i The Jaccard similarity coefficient, TD(c k ,x i ) means c k and x i The cosine similarity of α1, α2, α3 and α4 are distance weight coefficients, satisfying α1+α2+α 3+ α4=1, the improved K-means algorithm adopts dynamic update of cluster centers, and its update formula is:

[0011]

[0012] Among them, c k represents the kth cluster center, |c k | represents the number of cluster objects in the kth cluster, x i Indicates that the clustering object is a collection of temperature, current and power parameters;

[0013] S4. Perform distribution line fault diagnosis through the fault diagnosis model and transmit the results to the controller, which controls the alarm and communication system.

[0014] Preferably, in the S3 step, k in the K-means algorithm is 4, and any 4 power, temperature and current parameters within the same time period of a day are taken as the initial cluster centers, and the distance between the remaining data and the 12 data is calculated to determine the category to which the data belongs. If the data to be divided is empty, the average value of each category is calculated to determine the center point. If the data to be divided is not empty, the remaining data points are divided into the nearest clusters, and the new cluster center is re-determined. The distance between the remaining data and the cluster center is calculated again, and the data category is re-determined until the cluster center no longer changes or the preset number of iterations is reached.

[0015] Preferably, in the step S3, since the parameters are time-series, there will be N×4 clusters in N time periods, and the result with the highest degree of division is taken as the final classification, and the classification data is normalized.

[0016] Preferably, in step S3, the normalized parameters are input into the long short-term memory network and simplified into a one-dimensional sequence with time as the independent variable, the parameters of the long short-term memory network are set according to different electricity consumption types, and then data training is performed.

[0017] Preferably, the forget gate compares the previous state with the current input, and assigns a value of 0 or 1 to the information in the previous state to decide which information to retain or discard from the previous state, 1 means retaining the information, and 0 means discarding the information. The input gate stores the new information in the current state, and finds the relationship between current, power and temperature through the hidden layer. The fault is diagnosed according to the set parameters, and the information is output through the output gate. If it exceeds the set value, it is diagnosed as a fault, and if it does not exceed the set value, it is diagnosed as no fault.

[0018] Preferably, in step S4, the fault judgment result of the long short-term memory network is transmitted to the controller, and the controller is linked to the alarm communication system. If the result is a fault, the controller will activate the alarm and communication system.

[0019] Compared with the prior art, the advantages and positive effects of the present invention are:

[0020] 1. Adding a Savitzky-Go-Lay filter before the data is input into the improved long short-term memory network fault diagnosis model can remove outliers, noise and unreliable data, and can effectively smooth the signal data.

[0021] 2. The improved K-means algorithm is introduced into the long short-term memory network (LSTM) to classify the power distribution parameters into 4 categories, which improves the accuracy of subsequent fault judgment and prediction.

[0022] 3. The three gating mechanisms of the forget gate, input gate, and output gate inside the long-term and short-term memory network neurons will make judgments on the information at each moment and adjust and update it in a timely manner, conduct a comprehensive and dynamic analysis and evaluation of the entire distribution operation status, and promptly discover and alarm the existing fault hazards, so that accidents can be quickly eliminated automatically or manually. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0024] Figure 1 A schematic diagram of the structure of an intelligent power distribution monitoring and early warning method provided in an embodiment; DETAILED DESCRIPTION

[0025] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described below in conjunction with the accompanying drawings and embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0026] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways than those described herein. Therefore, the present invention is not limited to the specific embodiments of the following disclosure.

[0027] Examples, such as Figure 1 As shown, the present invention provides a method for intelligent power distribution monitoring and early warning.

[0028] First, perform a system self-test and obtain the temperature, current, and power parameters of all distribution cables with time-series characteristics.

[0029] Secondly, the Savitzky-Go-Lay filter is used to remove outliers, noise, and unreliable data, eliminate conflicts between time series data and environmental interference, and has good robustness in the presence of bad data.

[0030] Considering the power distribution link, the conventional recurrent neural network algorithm is currently used directly for fault diagnosis, without considering the different power distribution parameters and cable fault thresholds for different power consumption modes. Therefore, a lot of manpower and material costs will be wasted in the training and debugging link, and diagnostic errors are prone to occur in the fault diagnosis link, resulting in unnecessary maintenance and intervention. Therefore, in order to simplify the above process, avoid the time-consuming and tedious on-site parameter debugging link, and reduce the required manpower and material costs, an improved long short-term memory network fault diagnosis model is built, and an improved K-means algorithm is introduced before the data is input into the long short-term memory network:

[0031] A fault diagnosis model is built based on the long short-term memory network. Considering that different power consumption modes use different line parameters, an improved K-means algorithm is introduced before the long short-term memory network to divide power distribution into four categories: industrial power, agricultural power, commercial power, and municipal life power.

[0032] Considering that although the Euclidean distance can characterize the curve similarity relationship to a certain extent, it has certain limitations for parameters with time series characteristics. In order to eliminate the disadvantages of the Euclidean distance and take the morphological similarity of the parameters into consideration in the clustering process, the Fréchet distance is introduced to measure the similarity of the two sets. The Jaccard similarity coefficient is obtained by calculating the ratio of the intersection size to the union size of the two sets. Cosine similarity is used to measure the similarity of two vectors in direction, but it can also be applied to discrete data. It does not consider the size of the vector, but only the direction. Therefore, considering the similarity in various aspects, the improved K-means algorithm adds Fréchet distance, Jaccard similarity coefficient and cosine similarity to the Euclidean distance to construct a comprehensive distance. The formula for calculating the shortest distance between each clustering object and the cluster center is:

[0033] D(c k ,x i )=min{α1ED(c k ,x i )+α2FD(c k ,x i )+α3JD(c k ,x i )+α4TD(c k ,x i )}

[0034] Among them, c k represents the kth cluster center, x i Indicates that the clustering object is a collection of temperature, current and power parameters, ED(c k ,x i ) means c k and x i The Euclidean distance, FD(c k ,xi ) means c k and x i The Fréchet distance, JD(c k ,x i ) means c k and x i The Jaccard similarity coefficient, TD(c k ,x i ) means c k and x i The cosine similarity of α1, α2, α3 and α4 are distance weight coefficients, satisfying α1+α2+α 3+ α4=1.

[0035] Considering that the random selection of the initial cluster center makes the algorithm extremely easy to fall into the local optimal solution, the inter-class distance and the intra-class distance are combined to dynamically update the cluster center, which can ensure that the cluster centers of different classes are mutually exclusive as much as possible, thereby ensuring low similarity between classes and low similarity within classes, and ensuring the final accuracy of clustering. Therefore, the improved K-means algorithm adopts dynamic update of cluster centers, and its update formula is:

[0036]

[0037] Among them, c k represents the kth cluster center, |c k | represents the number of cluster objects in the kth cluster, x i Indicates that the clustering object is a collection of temperature, current and power parameters;

[0038] Specifically, k in the K-means algorithm is 4, and any four power, temperature and current parameters in the same time period of a day are used as the initial cluster centers. The distance between the remaining data and these 12 data is calculated to determine the data category. If the data to be divided is empty, the average value of each category is calculated to determine the center point. If the data to be divided is not empty, the remaining data points are divided into the nearest cluster, and the new cluster center is re-determined. The distance between the remaining data and these four data is calculated again, and the data category is re-determined until the cluster center no longer changes or the preset number of iterations is reached.

[0039] Taking into account the wide variety of power distribution cables, involving different voltage levels, current capacity, insulation materials, voltage resistance and other characteristics, the results obtained by the K-means algorithm are statistically analyzed again, and N time periods are divided into N×4 clusters. The results with the highest degree of similarity are taken as the final classification to improve the accuracy of classification.

[0040] Next, considering that traditional recurrent neural networks cannot classify, process and predict time series data and may encounter the problem of gradient disappearance during training, LSTM is relatively insensitive to the length of intervals between sequences, which is an advantage in many applications compared to RNN, hidden Markov model and other sequence learning methods. As a nonlinear model, LSTM can be used as a complex nonlinear unit to construct large deep neural networks.

[0041] Then, the normalized parameters are input into the long short-term memory network and simplified into a one-dimensional sequence with time as the independent variable. The parameters of the long short-term memory network are set according to different types of electricity consumption. Then, the forget gate compares the previous state with the current input, and assigns a value of 0 or 1 to the information in the previous state to decide which information to retain or discard from the previous state. 1 means retaining the information, and 0 means discarding the information. The input gate stores the new information in the current state, and finds the relationship between current, power and temperature through the hidden layer. The fault is diagnosed according to the set parameters, and the information is output through the output gate. If it exceeds the set value, it is diagnosed as a fault. If it does not exceed the set value, it is diagnosed as no fault. The long short-term memory network can comprehensively, accurately and in real time collect relevant information, conduct a comprehensive and dynamic analysis and evaluation of the entire distribution operation status, and timely discover and alarm the existing fault hazards, so that the accidents that occur can be eliminated quickly and automatically or manually.

[0042] Finally, the fault judgment result of the long short-term memory network is transmitted to the controller, and the controller is linked to the alarm communication system. If the result is a fault, the controller will activate the alarm and communication system.

[0043] The above description is only a preferred embodiment of the present invention and does not limit the present invention in other forms. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A method for intelligent power distribution monitoring and early warning, characterized in that: The following steps are involved: S1. The system self-checks and obtains the temperature, current and power parameters with time sequence characteristics of all distribution lines; S2, remove outliers, noise and unreliable data through Savitzky-Go-Lay filter, and eliminate conflicts and environmental interference between time series data; S3. Build a fault diagnosis model based on the long short-term memory network. Considering that different power consumption modes use different line parameters, introduce an improved K-means algorithm before the long short-term memory network to divide power distribution into four categories: industrial power, agricultural power, commercial power, and municipal life power. The improved K-means algorithm adds Fréchet distance, Jaccard similarity coefficient, and cosine similarity to the Euclidean distance to construct a comprehensive distance. The formula for calculating the shortest distance between each cluster object and the cluster center is: D(c k ,x i )=min{α1ED(c k ,x i )+α2FD(c k ,x i )+α3JD(c k ,x i )+α4TD(c k ,x i )} Among them, c k represents the kth cluster center, x i Indicates that the clustering object is a collection of temperature, current and power parameters, ED(c k ,x i ) means c k and x i The Euclidean distance, FD(c k ,x i ) means c k and x i The Fréchet distance, JD(c k ,x i ) means c k and x i The Jaccard similarity coefficient, TD(c k ,x i ) means c k and x i The cosine similarity of , α1, α2, α3 and α4 are distance weight coefficients, satisfying α1+α2+α3+α4=1, The improved K-means algorithm adopts dynamic update of cluster centers, and its update formula is: Among them, c k represents the kth cluster center, |c k | represents the number of cluster objects in the kth cluster; S4. Perform distribution line fault diagnosis through the fault diagnosis model and transmit the results to the controller, which controls the alarm and communication system.

2. The method for intelligent power distribution monitoring and early warning according to claim 1 is characterized in that: In the step S3, k in the K-means algorithm is taken as 4, and any 4 power, temperature and current parameters in the same time period of a day are used as initial cluster centers. The distance between the remaining data and these 12 data is calculated to determine the category to which the data belongs. If the data to be divided is empty, the average value of each category is calculated to determine the cluster center. If the data to be divided is not empty, the remaining data points are divided into the nearest cluster, and the new cluster center is re-determined. The distance between the remaining data and these 4 data is calculated again, and the data category is re-determined until the cluster center no longer changes or the preset number of iterations is reached.

3. The method for intelligent power distribution monitoring and early warning according to claim 1 is characterized in that: In the step S3, since the parameters are time-series, there will be N×4 clusters in N time periods. The result with the highest degree of similarity is used as the final classification, and the classification data is normalized.

4. The method for intelligent power distribution monitoring and early warning according to claim 1 is characterized in that: In step S3 shown, the normalized parameters are input into the long short-term memory network and simplified into a one-dimensional sequence with time as the independent variable. The parameters of the long short-term memory network are set according to different types of electricity consumption, and then data training is performed.

5. The method for intelligent power distribution monitoring and early warning according to claim 4 is characterized in that: The forget gate compares the previous state with the current input and assigns a value of 0 or 1 to the information in the previous state to decide which information to retain or discard from the previous state. 1 means retaining the information and 0 means discarding the information. The input gate stores the new information in the current state and finds the relationship between current, power and temperature through the hidden layer. The fault is diagnosed according to the set parameters and the information is output through the output gate. If it exceeds the set value, it is diagnosed as a fault. If it does not exceed the set value, it is diagnosed as no fault.

6. The method for intelligent power distribution monitoring and early warning according to claim 1 is characterized in that: In the step S4, the fault judgment result of the long short-term memory network is transmitted to the controller, and the controller is linked to the alarm and communication system. If the result is a fault, the controller will activate the alarm and communication system.

Images