Power distribution network fault prediction system and method based on time series data mining

Abstract

The invention discloses a power distribution network fault prediction system and method based on time series data mining. The method comprises the following steps: dividing a power distribution network into R sub-regions; collecting time sequence power grid characteristic data of the sub-regions; the time sequence power grid characteristic data comprises power grid characteristic data and time attributes corresponding to the power grid characteristic data; inputting the time sequence power grid characteristic data of the rth sub-region into a pre-trained data prediction model corresponding to the rth sub-region to obtain predicted power grid characteristic data of the rth sub-region at a future moment; comparing and analyzing the predicted power grid characteristic data of the rth sub-region and a preset power grid characteristic data threshold value of the rth sub-region at the future T moment, and judging whether abnormal elements exist in the predicted power grid characteristic data or not; therefore, the fault prediction and recognition of each sub-region are more accurate, and the reliability and safety of power grid operation can be comprehensively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power grid monitoring, and in particular to a distribution network fault prediction system and method based on time series data mining. Background Art

[0002] In modern power systems, the stability and reliability of distribution networks are of vital importance. With the continuous development of social economy and the rapid growth of people's demand for electricity, the load level of the power grid has increased significantly, and the power grid structure and operating environment have become increasingly complex. Especially in the face of variable power demand and diverse load characteristics, the failure rate of distribution networks has increased significantly. Once a failure occurs in the distribution network, it may not only cause local or even large-scale power outages, but also damage the distribution network equipment. In severe cases, it may also cause major safety accidents, affecting the normal lives of residents and the continued operation of industry and commerce; therefore, how to ensure the safe and stable operation of the distribution network and quickly and accurately detect and predict failures has become a key challenge facing the power industry.

[0003] The prediction accuracy of the time series feature data model in the existing technology is insufficient, and it cannot accurately capture the changing trend of the power grid characteristic data. It also does not consider the fault correlation between different sub-areas of the distribution network. The fault identification is not precise enough, which reduces the accuracy of fault prediction and the sensitivity of identification.

[0004] Therefore, it is an urgent problem for those skilled in the art to propose a distribution network fault prediction system and method based on time series data mining to solve the difficulties existing in the prior art. Summary of the invention

[0005] The purpose of the present invention is to provide a distribution network fault prediction system and method based on time series data mining to solve the shortcomings of the background technology.

[0006] In order to achieve the above object, the present invention provides the following technical solution: a distribution network fault prediction method based on time series data mining, comprising:

[0007] The distribution network is divided into R sub-areas, where R is an integer greater than 1;

[0008] Collecting time-series power grid characteristic data of the sub-area; the time-series power grid characteristic data includes the power grid characteristic data and the time attributes corresponding to the power grid characteristic data;

[0009] Input the time series grid characteristic data of the rth sub-region into the pre-trained data prediction model corresponding to the rth sub-region to obtain the predicted grid characteristic data of the rth sub-region at the future time, r∈R;

[0010] The predicted grid characteristic data of the r-th sub-region is compared and analyzed with the grid characteristic data threshold preset in the r-th sub-region at the future time T to determine whether there are abnormal elements in the predicted grid characteristic data.

[0011] If there are abnormal elements in the predicted power grid characteristic data, the fault form of the rth sub-area is identified, and the fault forms include self-generated faults and input faults;

[0012] If the fault form of the rth sub-region has a self-contained fault, the fault type of the rth sub-region is determined and warning information is generated.

[0013] Furthermore, the method for determining the fault type of the r-th sub-region includes:

[0014] Step 1001: Analyze the operating parameters of all distribution network equipment in the sub-area at the current moment to determine whether there is an abnormality in the operating parameters of the distribution network equipment. If there is no abnormality, identify the fault form as an input fault; if there is an abnormality, execute step 1002;

[0015] Step 1002: Obtain the specific name of the distribution network device having abnormal operating parameters, and the name of the abnormal operating parameter;

[0016] Step 1003: input the name of the abnormal element, the name of the abnormal operating parameter and the specific name of the distribution network equipment into the pre-built correlation analysis model, and output the correlation result, which includes correlation and non-correlation; execute step 1004;

[0017] Step 1004 , if the correlation result is correlated, then identify that the fault type of the r th sub-region is a self-generated fault; if the correlation result is uncorrelated, then identify that the fault type of the r th sub-region is a self-generated fault and an input fault.

[0018] Furthermore, the method for determining the fault type of the rth sub-area includes:

[0019] If the fault form only has its own fault, the abnormal element data existing in the predicted power grid characteristic data is extracted, and the abnormal element data is substituted into the fixed input data to form fault characteristic data, and the fault characteristic data is input into the pre-trained fault recognition model to obtain the fault type.

[0020] The fault characteristic data includes abnormal element data and normal element data in the predicted power grid characteristic data replaced by 0;

[0021] If the fault type is a self-owned fault or an input fault, the abnormal operating parameter name and the specific name of the distribution network equipment will be used as the fault type.

[0022] Furthermore, the time attribute includes a time attribute value and a time point attribute value, and the time attribute acquisition method includes:

[0023] Step S101: Obtain the time interval of the future time T in the preset time interval table, and use the time point attribute value corresponding to the time interval as the time point attribute value of the future time T; if the time of the future time T is a working day, execute step S102; if the time of the future time T is a non-working day, execute step S103;

[0024] Step S102: the time attribute value is A1;

[0025] Step S103: If the label date is a weekend, the time attribute value is marked as A2, and A1 is less than A2; if the label date is a holiday, the time attribute value is marked according to the number of days of the holiday. The longer the holiday, the larger the time attribute value.

[0026] Furthermore, the training method of the data prediction model includes:

[0027] Pre-collect a data sample set of the r-th sub-area, the data sample includes time-series power grid characteristic data, and convert the time-series power grid characteristic data into a feature vector; preset a prediction time step T, a sliding step, and a sliding window length; convert the data sample set into multiple training samples using a sliding window method, one training sample corresponds to one label, and constitutes a set of training data;

[0028] The training data is used as the input of the data prediction model, and the predicted power grid characteristic data of the rth sub-region at the future moment after the prediction time step T is used as the output. The actual power grid characteristic data of the rth sub-region at the future moment after each time step T is used as the prediction target, and the prediction accuracy is used as the training target. The data prediction model is trained to obtain a data prediction model that meets the prediction accuracy. The data prediction model is a long short-term memory network or a random forest regression model.

[0029] Furthermore, the method for determining whether there are abnormal elements in the predicted power grid characteristic data includes:

[0030] If any element in the predicted power grid characteristic data is greater than the corresponding threshold in the power grid characteristic data threshold, it is determined that there is an abnormal element in the predicted power grid characteristic data; otherwise, there is no abnormal element.

[0031] Furthermore, the method for presetting the threshold of the power grid characteristic data includes:

[0032] The time-series power grid characteristic data is input into the pre-trained threshold setting model corresponding to the r-th sub-region to obtain the power grid characteristic data threshold predicted for the r-th sub-region at the future time T.

[0033] Furthermore, the method of analyzing the operating parameters of all distribution network equipment in the sub-area at the current moment and determining whether there is an abnormality in the operating parameters of the distribution network equipment includes:

[0034] The operating parameters of all distribution network equipment at the current moment are compared and analyzed with the rated operating parameters of the corresponding distribution network equipment in turn. If the operating parameters of any distribution network equipment at the current moment are not within the rated operating parameter range of the corresponding distribution network equipment, it is determined that there is an abnormality in the operating parameters of the power grid equipment; otherwise, the opposite is true.

[0035] Furthermore, the training method of the correlation analysis model includes:

[0036] A correlation analysis data set is collected in advance, wherein the correlation analysis data includes correlation feature data and correlation results corresponding to the correlation feature data; the correlation feature data includes the name of the abnormal element, the name of the abnormal operating parameter and the specific name of the distribution network equipment; the correlation analysis data set is divided into a training set and a test set, the correlation feature data in the training set is used as the input of the correlation analysis model, and the correlation results in the training set are used as the output of the correlation analysis model, with minimizing the sum of the prediction accuracies of all predicted correlation results as the training goal; the training is stopped when the sum of the prediction accuracies converges; the correlation analysis model is a gradient boosting tree model.

[0037] A distribution network fault prediction system based on time series data mining, which implements the distribution network fault prediction method based on time series data mining, comprises:

[0038] A region division module is used to divide the distribution network into R sub-regions, where R is an integer greater than 1;

[0039] A collection module, used to collect time-series power grid characteristic data of the sub-area; the time-series power grid characteristic data includes the power grid characteristic data and the time attributes corresponding to the power grid characteristic data;

[0040] The first analysis module is used to input the time series power grid characteristic data of the rth sub-region into the pre-trained data prediction model corresponding to the rth sub-region to obtain the predicted power grid characteristic data of the rth sub-region at a future time, r∈R;

[0041] The second analysis module compares and analyzes the predicted grid characteristic data of the r-th sub-region with the grid characteristic data threshold preset for the r-th sub-region at the future time T to determine whether there are abnormal elements in the predicted grid characteristic data.

[0042] The third analysis module identifies the fault form of the rth sub-area if there are abnormal elements in the predicted power grid characteristic data, and the fault form includes self-generated fault and input fault;

[0043] The fourth analysis module determines the fault type of the rth sub-region and generates warning information if the fault form of the rth sub-region has a self-contained fault.

[0044] The present invention provides a distribution network fault prediction system and method based on time series data mining, and provides the following technical effects and advantages:

[0045] By collecting the time-series grid characteristic data of each sub-area and combining the grid characteristic data with the time attribute, the grid characteristic data and its corresponding threshold value at the future time T are predicted respectively. The set grid characteristic data threshold value can be adjusted according to the dynamic changes of the time-series grid characteristic data, that is, different grid characteristic data threshold values ​​are matched under different grid characteristic data and time attributes. By comparing and analyzing the actual grid characteristic data with the dynamic threshold value, the sensitivity and accuracy of anomaly detection can be effectively improved, and potential abnormal changes can be captured in time.

[0046] When abnormal elements appear in the predicted grid characteristic data, the system will identify the fault type of the rth sub-area. The fault forms include self-generated faults and input faults.

[0047] Specifically, the operating parameters of all distribution network equipment in the sub-area at the current moment are first analyzed to determine whether these parameters are abnormal. If no operating parameter abnormality is found, the fault is identified as an input fault, indicating that the fault comes from external input. If there is an operating parameter abnormality, the correlation between the name of the abnormal element, the name of the abnormal operating parameter, and the name of the distribution network equipment is further analyzed. According to the existence or non-existence of these correlations, based on such in-depth analysis, the system can more accurately locate the source of the fault and determine the corresponding fault type.

[0048] The advantage of this method is that it not only takes into account the dynamic changes and time dependence of power grid operation data, but also provides a more sophisticated fault identification mechanism through correlation analysis between abnormal elements and equipment operation status. Through this refined fault identification process, the accuracy and sensitivity of sub-region fault prediction can be greatly improved. Not only can the source of the problem be quickly located to prevent further expansion of the fault, but also false alarms or missed alarms can be reduced, and the operation and maintenance efficiency of the distribution network can be optimized. Ultimately, this fault identification mechanism based on dynamic threshold setting and correlation analysis makes fault prediction and identification in each sub-region more accurate, which helps to comprehensively improve the reliability and safety of power grid operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0050] Figure 1 It is a schematic diagram of a distribution network fault prediction system based on time series data mining of the present invention;

[0051] Figure 2 This is a flow chart of the distribution network fault prediction method based on time series data mining of the present invention. DETAILED DESCRIPTION

[0052] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0053] It should be noted that when an element is referred to as being "fixed to" another element, it may be directly on the other element or there may be an intermediate element. When an element is considered to be "connected to" another element, it may be directly connected to the other element or there may be an intermediate element at the same time.

[0054] Example 1

[0055] like Figure 1 As shown, the distribution network fault prediction system and method based on time series data mining in this embodiment include, through wired and / or wireless connection.

[0056] The regional division module is used to divide the distribution network into R sub-regions, where R is an integer greater than 1. The division can be based on the nature of electricity consumption in the sub-region, such as factory electricity consumption, supermarket electricity consumption, and residential electricity consumption.

[0057] The acquisition module is used to collect the time-series grid characteristic data of the sub-area; the time-series grid characteristic data includes the grid characteristic data and the time attributes corresponding to the grid characteristic data; the elements of the grid characteristic data include current, voltage, load, power factor, load balance and zero-sequence current. By monitoring the current, voltage, load, power factor, load balance and zero-sequence current of the distribution network, the overload fault, short circuit fault, ground fault, three-phase imbalance, low power factor fault, overvoltage fault and undervoltage fault of the distribution network can be analyzed; for example, by monitoring the zero-sequence current in the distribution network through a zero-sequence current transformer, the ground leakage fault in the power grid can be found. When the ground leakage fault occurs, the zero-sequence current will increase significantly.

[0058] The time attribute includes a time attribute value and a time point attribute value, and the time attribute acquisition method includes:

[0059] Step S101: Obtain the time interval of the future time T in the preset time interval table, and use the time point attribute value corresponding to the time interval as the time point attribute value of the future time T; if the time of the future time T is a working day, execute step S102; if the time of the future time T is a non-working day, execute step S103;

[0060] Step S102: the time attribute value is A1.

[0061] Step S103: If the label date is a weekend, the time attribute value is marked as A2, and A1 is less than A2; if the label date is a holiday, the time attribute value is marked according to the number of days of the holiday. The longer the holiday, the larger the time attribute value.

[0062] It should be noted that both the time attribute value and the time point attribute value will affect the grid characteristic data; for example, if the rth sub-area is a residential area, the time attribute value is A1, the time interval corresponding to the time point attribute value is the working time, the time attribute value and the time point attribute value are the working time on weekdays, and people go from the residential area to the office area to work, then the current and load in the distribution network of the rth sub-area will decrease; the voltage will rise due to the load reduction; the electrical equipment in the rth sub-area working time is mostly low-power inductive loads (such as refrigerators, standby equipment, etc.), these equipment will cause the power factor to decrease; the zero-sequence current will decrease as the load decreases, the three-phase current tends to balance. When other sub-areas are office areas, the changes in grid characteristic data are opposite.

[0063] The first analysis module is used to input the time series power grid characteristic data of the rth sub-region into the pre-trained data prediction model corresponding to the rth sub-region to obtain the predicted power grid characteristic data of the rth sub-region at a future time, r∈R;

[0064] The training methods of data prediction models include:

[0065] Collect the data sample set of the rth sub-area in advance, the data sample includes the time series power grid characteristic data, and convert the time series power grid characteristic data into a feature vector. Preset the prediction time step T, sliding step and sliding window length; use the sliding window method to convert the data sample set into multiple training samples, one training sample corresponds to one label, and constitute a set of training data.

[0066] The training data is used as the input of the data prediction model, and the predicted power grid characteristic data of the rth sub-region at the future moment after the prediction time step T is used as the output. The actual power grid characteristic data of the rth sub-region at the future moment after each time step T is used as the prediction target, and the prediction accuracy is used as the training target. The data prediction model is trained to obtain a data prediction model that meets the prediction accuracy. The data prediction model is a long short-term memory network (LSTM) or a random forest regression model.

[0067] The second analysis module compares and analyzes the predicted grid characteristic data of the r-th sub-region with the grid characteristic data threshold preset for the r-th sub-region at the future time T to determine whether there are abnormal elements in the predicted grid characteristic data.

[0068] Methods for determining whether there are abnormal elements in the predicted power grid characteristic data include:

[0069] The grid characteristic data thresholds include current threshold, voltage threshold, load threshold, power factor threshold, load balance threshold and zero-sequence current threshold;

[0070] If any element in the predicted power grid characteristic data is greater than the corresponding threshold in the power grid characteristic data threshold, it is determined that there is an abnormal element in the predicted power grid characteristic data; otherwise, it does not exist; any element in the predicted power grid characteristic data is current, voltage, load, power factor, load balance or zero-sequence current.

[0071] The preset methods of the power grid characteristic data thresholds include:

[0072] The time-series power grid characteristic data is input into the threshold setting model corresponding to the pre-trained r-th sub-region to obtain the power grid characteristic data threshold predicted for the r-th sub-region at the future time T.

[0073] The power grid characteristic data threshold set in this way can be dynamically set as the time-series power grid characteristic data changes. That is, under different power grid characteristic data and different time attributes, the corresponding power grid characteristic data threshold is also different, ensuring the sensitivity and accuracy of obtaining abnormal elements.

[0074] The training methods for the threshold setting model include:

[0075] The data sample set of the rth sub-area is collected in advance, and the data sample includes time-series power grid characteristic data, and the time-series power grid characteristic data is converted into a feature vector. The prediction time step T, sliding step and sliding window length are preset; the data sample set is converted into multiple training samples using the sliding window method, and one training sample corresponds to one label, and constitutes a set of training data.

[0076] The training data is used as the input of the data prediction model, and the predicted power grid characteristic data threshold of the rth sub-region at the future moment after the prediction time step T is used as the output. The actual power grid characteristic data threshold of the rth sub-region at the future moment after each time step T is used as the prediction target, and the prediction accuracy is used as the training target. The data prediction model is trained to obtain a data prediction model that meets the prediction accuracy. The data prediction model is a long short-term memory network (LSTM) or a random forest regression model.

[0077] The third analysis module identifies the fault type of the r-th sub-area if there are abnormal elements in the predicted power grid characteristic data. The fault types include self-generated faults and input faults. Self-generated faults are the faults predicted this time caused by the distribution network equipment in the r-th sub-area; input faults are faults caused by the distribution network equipment in other sub-areas, which affect the abnormality of the power grid characteristic data in the r-th sub-area, thereby causing the predicted fault. If they are not distinguished, the source of the predicted fault cannot be accurately solved, and when the future moment arrives, it will still cause a distribution network fault.

[0078] Methods for determining the fault form of the rth sub-area include:

[0079] Step 1001: Analyze the operating parameters of all distribution network equipment in the sub-area at the current moment to determine whether there is an abnormality in the operating parameters of the distribution network equipment. If there is no abnormality, the fault form is identified as an input fault, indicating that all distribution network equipment in the r-th sub-area at the current moment are operating normally. The predicted fault of the distribution network at the future time T is the r-th sub-area power grid fault caused by the distribution network fault in other sub-areas. The source of the fault is not in the r-th sub-area, so there is no need to initiate maintenance on the distribution network equipment in the r-th sub-area. If there is an abnormality, execute step 1002.

[0080] Step 1002: Obtain the specific name of the distribution network device having abnormal operating parameters, and the name of the abnormal operating parameter;

[0081] Step 1003, input the name of the abnormal element, the name of the abnormal operating parameter and the specific name of the distribution network equipment into the pre-built correlation analysis model, and output the correlation result, which includes correlation and irrelevance; execute step 1004; it should be noted that the number of the name of the abnormal element, the name of the abnormal operating parameter and the specific name of the distribution network equipment in the input data of the correlation analysis model is one. When there are multiple names of the abnormal elements and the names of the abnormal operating parameters, they need to be input one by one into the correlation analysis model for predictive analysis to ensure the consistency of the input data of the correlation analysis model and improve the prediction accuracy.

[0082] Step 1004, if the correlation result is correlated, the fault type of the r-th sub-area is identified as a self-generated fault; if the correlation result is uncorrelated, the fault type of the r-th sub-area is identified as a self-generated fault and an input fault, that is, both fault types exist, which means that the input fault (i.e., caused by the fault of the distribution network equipment in other sub-areas) causes the fault in the r-th sub-area. If the self-generated fault in the r-th sub-area is not handled in time, and the distribution network equipment in other sub-areas is not effectively maintained, then at the future time T, the distribution network equipment in the r-th sub-area will also be damaged; for example, when the input fault is an overload fault or a grounding fault, when the self-generated fault in the r-th sub-area identified at this time is a circuit breaker fault, if the circuit breaker fault is not maintained in time, it will most likely cause irreversible damage to the distribution network equipment in the r-th sub-area.

[0083] The method for analyzing the operating parameters of all distribution network equipment in the sub-area at the current moment and determining whether the operating parameters of the distribution network equipment are abnormal includes:

[0084] The operating parameters of all distribution network equipment at the current moment are compared and analyzed with the rated operating parameters of the corresponding distribution network equipment in turn. If the operating parameters of any distribution network equipment at the current moment are not within the rated operating parameter range of the corresponding distribution network equipment, it is determined that there is an abnormality in the operating parameters of the power grid equipment; otherwise, the opposite is true; for example, when the distribution network equipment is a transformer, the rated operating parameters of the transformer include rated current, rated capacity, rated load rate, rated noise range, etc.

[0085] The training methods of the correlation analysis model include:

[0086] A correlation analysis data set is collected in advance, wherein the correlation analysis data includes correlation feature data and correlation results corresponding to the correlation feature data; the correlation feature data includes the name of the abnormal element, the name of the abnormal operating parameter and the specific name of the distribution network equipment; the correlation analysis data set is divided into a training set and a test set, the correlation feature data in the training set is used as the input of the correlation analysis model, and the correlation results in the training set are used as the output of the correlation analysis model, with minimizing the sum of the prediction accuracies of all predicted correlation results as the training goal; the training is stopped when the sum of the prediction accuracies converges; the correlation analysis model is a gradient boosting tree model.

[0087] The calculation formula for prediction accuracy is: n =(Y n -J n ) 2 , where Z n is the prediction accuracy, Y n is the predicted correlation result corresponding to the nth group of correlation feature data, SJ n It is the actual correlation result corresponding to the nth group of correlation feature data.

[0088] The third analysis module determines the fault type of the rth sub-region if the fault form of the rth sub-region has a self-contained fault, generates warning information, and sends it to the workbench for review by the staff.

[0089] The method for determining the fault type of the rth sub-area includes:

[0090] If the fault form only has its own fault, the abnormal element data existing in the predicted power grid characteristic data is extracted, and the abnormal element data is substituted into the fixed input data to form fault characteristic data, and the fault characteristic data is input into the pre-trained fault recognition model to obtain the fault type.

[0091] The fault characteristic data includes abnormal element data and normal element data in the predicted power grid characteristic data replaced by 0; this maintains the consistency of the input data of the fault identification model each time, ensuring that the fault identification model can operate normally.

[0092] If the fault type is a self-owned fault or an input fault, the abnormal operating parameter name and the specific name of the distribution network equipment will be used as the fault type.

[0093] Fault types include overload fault, ground fault, overvoltage and undervoltage fault, etc.; in case of overload fault, the continuous load is large, the current increases and the power factor decreases; in case of ground fault, the zero-sequence current will increase significantly, and in case of single-phase grounding, the voltage of the faulty phase will decrease, and the voltage of other phases may increase, and the load balance will increase, that is, the three-phase imbalance will increase.

[0094] The training methods of the fault identification model include:

[0095] A fault identification data set is collected in advance, wherein the fault identification data includes fault feature data and the fault type corresponding to the fault feature data; the fault identification data set is divided into a training set and a test set, the fault feature data in the training set is used as the input of the fault identification model, and the fault type in the training set is used as the output of the fault identification model, with minimizing the sum of the prediction accuracies of all predicted fault types as the training goal; the training is stopped when the sum of the prediction accuracies reaches convergence; the fault identification model is a gradient boosting tree model.

[0096] By collecting the time-series grid characteristic data of each sub-area and combining the grid characteristic data with the time attribute, the grid characteristic data and its corresponding threshold value at the future time T are predicted respectively. The set grid characteristic data threshold value can be adjusted according to the dynamic changes of the time-series grid characteristic data, that is, different grid characteristic data threshold values ​​are matched under different grid characteristic data and time attributes. By comparing and analyzing the actual grid characteristic data with the dynamic threshold value, the sensitivity and accuracy of anomaly detection can be effectively improved, and potential abnormal changes can be captured in time.

[0097] When abnormal elements appear in the predicted grid characteristic data, the system will identify the fault type of the rth sub-area. The fault forms include self-generated faults and input faults.

[0098] Specifically, the operating parameters of all distribution network equipment in the sub-area at the current moment are first analyzed to determine whether these parameters are abnormal. If no operating parameter abnormality is found, the fault is identified as an input fault, indicating that the fault comes from external input. If there is an operating parameter abnormality, the correlation between the name of the abnormal element, the name of the abnormal operating parameter, and the name of the distribution network equipment is further analyzed. According to the existence or non-existence of these correlations, based on such in-depth analysis, the system can more accurately locate the source of the fault and determine the corresponding fault type.

[0099] The advantage of this method is that it not only takes into account the dynamic changes and time dependence of power grid operation data, but also provides a more sophisticated fault identification mechanism through correlation analysis between abnormal elements and equipment operation status. Through this refined fault identification process, the accuracy and sensitivity of sub-area fault prediction can be greatly improved. Not only can the source of the problem be quickly located to prevent further expansion of the fault, but also false alarms or missed alarms can be reduced, and the operation and maintenance efficiency of the distribution network can be optimized. Ultimately, this fault identification mechanism based on dynamic threshold setting and correlation analysis makes fault prediction and identification in each sub-area more accurate, which helps to comprehensively improve the reliability and safety of power grid operation.

[0100] Example 2

[0101] See also Figure 2 As shown, this embodiment provides a distribution network fault prediction method based on time series data mining, including: a distribution network fault prediction method based on time series data mining, including:

[0102] The distribution network is divided into R sub-areas, where R is an integer greater than 1;

[0103] Collecting time-series power grid characteristic data of the sub-area; the time-series power grid characteristic data includes the power grid characteristic data and the time attributes corresponding to the power grid characteristic data;

[0104] Input the time series grid characteristic data of the rth sub-region into the pre-trained data prediction model corresponding to the rth sub-region to obtain the predicted grid characteristic data of the rth sub-region at the future time, r∈R;

[0105] The predicted grid characteristic data of the r-th sub-region is compared and analyzed with the grid characteristic data threshold preset in the r-th sub-region at the future time T to determine whether there are abnormal elements in the predicted grid characteristic data.

[0106] If there are abnormal elements in the predicted power grid characteristic data, the fault form of the rth sub-area is identified, and the fault forms include self-generated faults and input faults;

[0107] If the fault form of the rth sub-region has a self-contained fault, the fault type of the rth sub-region is determined and warning information is generated.

[0108] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A distribution network fault prediction method based on time series data mining, characterized in that: include: The distribution network is divided into R sub-areas, where R is an integer greater than 1; Collecting time-series grid characteristic data of sub-regions; Timing The power grid characteristic data includes the power grid characteristic data and a time attribute corresponding to the power grid characteristic data; Input the time series grid characteristic data of the rth sub-region into the pre-trained data prediction model corresponding to the rth sub-region to obtain the predicted grid characteristic data of the rth sub-region at the future time, r∈R; Compare and analyze the predicted grid characteristic data of the r-th sub-region with the grid characteristic data threshold preset at the r-th sub-region at the future time T to determine whether there are abnormal elements in the predicted grid characteristic data; If there are abnormal elements in the predicted power grid characteristic data, the fault form of the rth sub-area is identified, and the fault forms include self-generated faults and input faults; If the fault form of the rth sub-region has a self-contained fault, the fault type of the rth sub-region is determined and warning information is generated.

2. The distribution network fault prediction method based on time series data mining according to claim 1 is characterized in that: Methods for determining the fault form of the rth sub-area include: Step 1001: Analyze the operating parameters of all distribution network equipment in the sub-area at the current moment to determine whether there is an abnormality in the operating parameters of the distribution network equipment. If there is no abnormality, identify the fault form as an input fault; if there is an abnormality, execute step 1002; Step 1002: Obtain the specific name of the distribution network device having abnormal operating parameters, and the name of the abnormal operating parameter; Step 1003: input the name of the abnormal element, the name of the abnormal operating parameter and the specific name of the distribution network equipment into the pre-built correlation analysis model, and output the correlation result, which includes correlation and non-correlation; execute step 1004; Step 1004: if the correlation result is correlated, then identify that the fault type of the r-th sub-region is a self-generated fault; if the correlation result is uncorrelated, then identify that the fault type of the r-th sub-region is a self-generated fault and an input fault.

3. The distribution network fault prediction method based on time series data mining according to claim 2 is characterized in that: The method for determining the fault type of the rth sub-area includes: If the fault form only has self-type faults, then the abnormal element data existing in the predicted power grid characteristic data is extracted, and the abnormal element data is substituted into the fixed input data to form fault characteristic data, and the fault characteristic data is input into the pre-trained fault recognition model to obtain the fault type; The fault characteristic data includes abnormal element data and normal element data in the predicted power grid characteristic data replaced by 0; If the fault type is a self-owned fault or an input fault, the abnormal operating parameter name and the specific name of the distribution network equipment will be used as the fault type.

4. The distribution network fault prediction method based on time series data mining according to claim 1 is characterized in that: The time attribute includes a time attribute value and a time point attribute value, and the time attribute acquisition method includes: Step S101: Obtain the time interval of the future time T in the preset time interval table, and use the time point attribute value corresponding to the time interval as the time point attribute value of the future time T; if the time of the future time T is a working day, execute step S102; if the time of the future time T is a non-working day, execute step S103; Step S102: the time attribute value is A1; Step S103: If the label date is a weekend, the time attribute value is marked as A2, and A1 is less than A2; if the label date is a holiday, the time attribute value is marked according to the number of days of the holiday. The longer the holiday, the larger the time attribute value.

5. The distribution network fault prediction method based on time series data mining according to claim 1 is characterized in that: The training method of the data prediction model includes: Pre-collect a data sample set of the r-th sub-area, the data sample includes time-series power grid characteristic data, and convert the time-series power grid characteristic data into a feature vector; preset a prediction time step T, a sliding step, and a sliding window length; convert the data sample set into multiple training samples using a sliding window method, one training sample corresponds to one label, and constitutes a set of training data; The training data is used as the input of the data prediction model, and the predicted power grid characteristic data of the rth sub-region at the future moment after the prediction time step T is used as the output. The actual power grid characteristic data of the rth sub-region at the future moment after each time step T is used as the prediction target, and the prediction accuracy is used as the training target. The data prediction model is trained to obtain a data prediction model that meets the prediction accuracy. The data prediction model is a long short-term memory network or a random forest regression model.

6. The distribution network fault prediction method based on time series data mining according to claim 1 is characterized in that: Methods for determining whether there are abnormal elements in the predicted power grid characteristic data include: If any element in the predicted power grid characteristic data is greater than the corresponding threshold in the power grid characteristic data threshold, it is determined that there is an abnormal element in the predicted power grid characteristic data; otherwise, there is no abnormal element.

7. The distribution network fault prediction method based on time series data mining according to claim 1 is characterized in that: The method for presetting the threshold value of the power grid characteristic data includes: The time-series power grid characteristic data is input into the pre-trained threshold setting model corresponding to the r-th sub-region to obtain the power grid characteristic data threshold predicted for the r-th sub-region at the future time T.

8. The distribution network fault prediction method based on time series data mining according to claim 1 is characterized in that: The method for analyzing the operating parameters of all distribution network equipment in the sub-area at the current moment and determining whether the operating parameters of the distribution network equipment are abnormal includes: The operating parameters of all distribution network equipment at the current moment are compared and analyzed with the rated operating parameters of the corresponding distribution network equipment in turn. If the operating parameters of any distribution network equipment at the current moment are not within the rated operating parameter range of the corresponding distribution network equipment, it is determined that there is an abnormality in the operating parameters of the power grid equipment; otherwise, the opposite is true.

9. The distribution network fault prediction method based on time series data mining according to claim 2 is characterized in that: The training method of the correlation analysis model includes: A correlation analysis data set is collected in advance, wherein the correlation analysis data includes correlation feature data and correlation results corresponding to the correlation feature data; the correlation feature data includes the name of the abnormal element, the name of the abnormal operating parameter and the specific name of the distribution network equipment; the correlation analysis data set is divided into a training set and a test set, the correlation feature data in the training set is used as the input of the correlation analysis model, and the correlation results in the training set are used as the output of the correlation analysis model, with minimizing the sum of the prediction accuracies of all predicted correlation results as the training goal; the training is stopped when the sum of the prediction accuracies converges; the correlation analysis model is a gradient boosting tree model.

10. A distribution network fault prediction system based on time series data mining, characterized in that: Implementing the distribution network fault prediction method based on time series data mining as described in any one of claims 1 to 9, comprising: A region division module is used to divide the distribution network into R sub-regions, where R is an integer greater than 1; A collection module, used to collect time-series power grid characteristic data of the sub-area; the time-series power grid characteristic data includes the power grid characteristic data and the time attributes corresponding to the power grid characteristic data; The first analysis module is used to input the time series power grid characteristic data of the rth sub-region into the pre-trained data prediction model corresponding to the rth sub-region to obtain the predicted power grid characteristic data of the rth sub-region at a future time, r∈R; The second analysis module compares and analyzes the predicted grid characteristic data of the r-th sub-region with the grid characteristic data threshold preset in the r-th sub-region at the future time T to determine whether there are abnormal elements in the predicted grid characteristic data; The third analysis module identifies the fault form of the rth sub-area if there are abnormal elements in the predicted power grid characteristic data, and the fault form includes self-generated fault and input fault; The fourth analysis module determines the fault type of the rth sub-region and generates warning information if the fault form of the rth sub-region has a self-contained fault.

Images