Node anomaly diagnosis method and system based on carrier time-frequency synchronization network cluster

Abstract

The application discloses a kind of based on carrier time-frequency synchronization network cluster node exception diagnosis method and system, the method includes: based on carrier time-frequency network acquisition time-frequency node real-time measurement information;The time-frequency node real-time measurement information is divided into characteristic quantity and abnormal type;From the time-frequency node real-time measurement information, a part of data is randomly selected as verification set, the rest part of data is as training set, learns a decision tree from training set based on decision tree generation algorithm C4.5;According to precision and recall rate, the performance of decision tree in verification set is evaluated, if performance cannot reach set threshold, the decision tree is pruned, the decision tree after pruning is retrained, until the performance of decision tree reaches set threshold or iteration number exceeds preset upper limit.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of metrology, and more particularly, to a method and system for diagnosing node abnormalities in a carrier-based time-frequency synchronization network cluster. BACKGROUND

[0002] Currently, power line carrier communication technology has been fully used in State Grid substations, and metering devices collect data and perform remote fee control and other operations through carrier communication. Actual operation data statistics show that the communication quality is stable and reliable, and the substation carrier communication technology has matured. In order to solve the problem that the current numerous metering devices in substations lack effective time synchronization means, the time accuracy error is large, and cannot meet the needs of new power systems, a time-frequency network based on carrier technology is currently being researched and constructed to perform online time-frequency value transmission and online time-frequency performance monitoring on metering devices in substations. However, there is no related research on real-time abnormal diagnosis technology for carrier time-frequency networks and time-frequency nodes. The present method proposes an abnormal diagnosis method based on big data according to the characteristics of time-frequency networks and time-frequency nodes, which plays a key role in effectively improving the operation efficiency of future carrier-based time-frequency value transmission networks. SUMMARY

[0003] According to the present application, a method and system for diagnosing node abnormalities in a carrier-based time-frequency synchronization network cluster are provided to solve the technical problems of the deficiencies of existing network abnormal diagnosis technology and the actual situation of using carrier communication for substation time-frequency synchronization.

[0004] According to a first aspect of the present application, a method for diagnosing node abnormalities in a carrier-based time-frequency synchronization network cluster is provided, comprising:

[0005] determining a carrier-based substation time-frequency synchronization network, and collecting real-time measurement information of time-frequency nodes based on the carrier time-frequency network;

[0006] dividing the real-time measurement information of the time-frequency nodes into feature quantities and abnormal types, extracting the feature quantities of the real-time measurement information of the time-frequency nodes, obtaining time-frequency node feature values, determining the range of abnormal nodes by analyzing the time-frequency node feature values, and dividing the abnormal types of the time-frequency nodes by analyzing the evolution law of single-node time-frequency parameters and the evolution law of adjacent nodes in the cluster;

[0007] randomly selecting a part of data from the real-time measurement information of the time-frequency nodes as a validation set, and the remaining data as a training set, learning a decision tree from the training set based on a decision tree generation algorithm C4.5;

[0008] evaluating the performance of the decision tree in the validation set according to the precision and recall rate, pruning the decision tree if the performance cannot reach the set threshold, retraining the pruned decision tree, and stopping until the performance of the decision tree reaches the set threshold or the number of iterations exceeds the preset upper limit.

[0009] Optionally, a carrier-based cell area time-frequency synchronization network is determined, and time-frequency node real-time measurement information is collected based on the carrier time-frequency network, including:

[0010] The cell area time-frequency synchronization network is determined to include a plurality of terminal nodes, i.e., one carrier master clock node and a plurality of carrier slave clock nodes;

[0011] The terminal nodes are established in a topological relationship based on the power carrier network, and a synchronization system is established in the entire power carrier network. In the topological relationship: the T-GM is the network master node and is the master clock; the T-TC is a relay node and is both a slave clock and a transparent clock; and the T-SC is an end node and is a slave clock.

[0012] The carrier is passed through the T-GM master node, the time-frequency synchronization of other nodes in the network is performed, and time-frequency node real-time measurement information is collected. The time-frequency node real-time measurement information includes: device number, superior device number, subordinate device number, frequency error, time error, network delay, last time-frequency synchronization time, and environmental temperature.

[0013] Optionally, a feature quantity of the time-frequency node real-time measurement information is extracted, and a time-frequency node feature value is obtained, including:

[0014] The time-frequency node feature value includes: frequency temperature correction coefficient, node temperature-corrected normalized frequency difference, node normalized frequency difference mean value, node normalized frequency difference standard deviation, node frequency deviation aging rate, node time difference mean value, node time difference standard deviation, frequency deviation between superior and subordinate nodes, time deviation between superior and subordinate nodes, frequency deviation between superior and subordinate nodes, time deviation between superior and subordinate nodes, node network delay mean value, and node network delay standard deviation.

[0015] According to the accurate frequency value fyi of the master clock, the frequency difference measurement value fci of the slave clock, and the real-time temperature measurement value Ti, the fitting is performed according to (fci / fyi)=k1*Ti+k2*Ti 2 +k3*Ti 3 , and the frequency temperature correction coefficient K (k1, k2, k3) is obtained online.

[0016] Based on the frequency difference measurement value fci and the temperature measurement value Ti, the master node frequency measurement value fyi, and the node temperature correction coefficient K, the node temperature-corrected normalized frequency difference fc is obtained according to fc=[fci-fyi*(1+k1*Ti+k2*Ti2+k3*Ti3)] / fyi.

[0017] The node temperature-corrected normalized frequency difference fc measurement results of the node within a period of time are averaged to obtain the node normalized frequency difference mean value fm.

[0018] The standard deviation of the normalized frequency difference fc of the node after multiple node temperature corrections in a period of time is obtained, and the normalized frequency difference standard deviation fs of the node is obtained;

[0019] The mean value fmi of the normalized frequency difference of the node after temperature correction in a period of time is fitted according to fmi = λ * T + c, and the node frequency deviation aging rate λ is obtained;

[0020] The mean value ΔTm of the time difference of the node is obtained by averaging the time difference measurement results of the node in a period of time;

[0021] The standard deviation ΔTs of the time difference of the node is obtained by taking the standard deviation of the time difference measurement results of the node in a period of time;

[0022] The frequency deviation fuc between the upper node is obtained by subtracting the normalized frequency difference of the upper node at the same time from the normalized frequency difference of the node;

[0023] The time deviation ΔTuc between the upper node is obtained by subtracting the time deviation of the upper node at the same time from the time deviation of the node;

[0024] The frequency deviation fnc between the lower node is obtained by subtracting the normalized frequency difference of the lower node at the same time from the normalized frequency difference of the node;

[0025] The time deviation ΔTnc between the lower node is obtained by subtracting the time deviation of the lower node at the same time from the time deviation of the node;

[0026] The mean value ΔTnm of the network delay of the node in a period of time is obtained by averaging the network delay measurement values of the node;

[0027] The standard deviation ΔTns of the network delay of the node in a period of time is obtained by calculating the standard deviation of the network delay measurement values of the node.

[0028] Optionally, the abnormal node range is determined by analyzing the time-frequency node characteristic value, and the time-frequency node abnormal type is divided by analyzing the single node time-frequency parameter evolution law and the evolution law of the time-frequency parameter of the adjacent node in the cluster, including:

[0029] The abnormal node range is determined by analyzing the time-frequency node characteristic value, and the time-frequency node abnormal type is divided into single node time-frequency synchronization data packet loss fault, single node synchronization network instability fault, single node time-frequency performance attenuation threshold fault, single node time-frequency performance mutation fault, regional time-frequency synchronization data packet loss fault, and regional synchronization network delay instability fault by analyzing the single node time-frequency parameter evolution law and the time-frequency parameter evolution law of the adjacent node in the cluster.

[0030] Single node time-frequency synchronization data packet loss fault refers to the node monitoring data that can be collected, but the node time is not updated, and the lower node time-frequency characteristics are consistent with the node, while the upper node time-frequency characteristics are normal.

[0031] Single node synchronization network delay instability fault refers to the node network delay standard deviation exceeding the threshold value.

[0032] Single node synchronization network instability fault refers to the node having a network offline state.

[0033] Single node time-frequency performance decay threshold fault refers to the node frequency deviation aging rate exceeding the threshold value, and the upper node time-frequency characteristics are normal.

[0034] Single node time-frequency performance mutation fault refers to the comprehensive evaluation of the node normalized frequency difference mean, the node normalized frequency difference standard deviation, the node time difference mean, and the node time difference standard deviation. The evaluation result exceeds the threshold value, and the upper node time-frequency characteristics are normal, and the lower node time-frequency performance is consistent with the node.

[0035] Regional time-frequency synchronization data packet loss fault refers to the existence of data packet loss fault in multiple nodes in the area.

[0036] Regional synchronization network delay instability fault refers to the existence of network delay instability fault in multiple nodes in the area.

[0037] Regional synchronization network instability fault refers to the existence of network offline state in multiple nodes in the area.

[0038] Time-frequency device batch quality fault refers to the existence of performance decay threshold fault or time-frequency performance mutation fault in multiple devices in the running device.

[0039] Optionally, the performance of the decision tree in the validation set is evaluated according to the precision and recall, and if the performance cannot reach the set threshold, the decision tree is pruned, and the pruned decision tree is retrained until the performance of the decision tree reaches the set threshold or the iteration number exceeds the preset upper limit, including:

[0040] According to the precision and recall, the performance of the decision tree in the validation set is evaluated, and if the performance cannot reach the set threshold, the hyperparameter probability distribution is defined to prune the decision tree, the Markov chain Monte Carlo chain is initialized, and the feature quantity of the real-time measurement information of the time-frequency node is considered to ensure that the pruning strategy is optimized for the actual application scenario.

[0041] The pruning strategy is sampled, and the theoretical performance of each pruning strategy on the validation set is evaluated, and the hyperparameter probability distribution is updated based on the evaluation performance, so that the pruning strategy is gradually optimized.

[0042] The pruned decision tree is retrained by the updated hyperparameters, and the decision tree is trained on the training set and its performance is evaluated on the validation set, and the above process is repeated until the performance of the decision tree reaches a set threshold or the number of iterations exceeds a preset upper limit.

[0043] According to another aspect of the application, a node anomaly diagnosis system based on a carrier time-frequency synchronization network cluster is also provided, comprising:

[0044] A real-time measurement information determination module is configured to determine a carrier-based substation time-frequency synchronization network and collect real-time measurement information of time-frequency nodes based on the carrier time-frequency network.

[0045] A real-time measurement information division module is configured to divide the real-time measurement information of the time-frequency nodes into feature quantities and abnormal types, extract feature quantities of the real-time measurement information of the time-frequency nodes, obtain time-frequency node feature values, determine an abnormal node range by analyzing the time-frequency node feature values, and divide the time-frequency node abnormal types by analyzing single-node time-frequency parameter evolution rules and neighboring node time-frequency parameter evolution rules within the cluster.

[0046] A decision tree learning module is configured to randomly select a part of data from the real-time measurement information of the time-frequency nodes as a validation set, and the remaining data as a training set, and learn a decision tree from the training set based on a decision tree generation algorithm C4.5.

[0047] A validation set performance evaluation module is configured to evaluate the performance of the decision tree on the validation set according to precision and recall rate, prune the decision tree if the performance cannot reach a set threshold, retrain the pruned decision tree until the performance of the decision tree reaches the set threshold or the number of iterations exceeds a preset upper limit.

[0048] Optionally, the real-time measurement information determination module comprises:

[0049] A time-frequency synchronization network determination submodule is configured to determine that the substation time-frequency synchronization network comprises a plurality of terminal nodes, i.e., one carrier master clock node and a plurality of carrier slave clock nodes.

[0050] A topology relationship establishment submodule is configured to establish a topology relationship among the terminal nodes based on the power carrier network, and further establish a synchronization system in the entire power carrier network, wherein in the topology relationship, the T-GM is a network master node and a master clock, the T-TC is a relay node and is both a slave clock and a transparent clock, and the T-SC is an end node and is a slave clock.

[0051] The time-frequency node real-time measurement information acquisition submodule is configured to acquire time-frequency node real-time measurement information through a T-GM master node, and synchronize time and frequency of other nodes in the network, wherein the time-frequency node real-time measurement information includes device number, superior device number, inferior device number, frequency error, time error, network delay, last time-frequency synchronization time and environment temperature.

[0052] Optionally, the real-time measurement information division module includes:

[0053] The time-frequency node characteristic value includes frequency temperature correction coefficient, node temperature-corrected normalized frequency difference, node normalized frequency difference mean value, node normalized frequency difference standard deviation, node frequency deviation aging rate, node time difference mean value, node time difference standard deviation, frequency deviation between the node and a superior node, time deviation between the node and the superior node, frequency deviation between the node and an inferior node, time deviation between the node and the inferior node, node network delay mean value and node network delay standard deviation.

[0054] The frequency temperature correction coefficient obtaining submodule is configured to obtain frequency temperature correction coefficient K (k1, k2, k3) according to accurate frequency value fyi of a master clock, frequency difference measurement value fci of a clock, real-time temperature measurement value Ti, and fitting according to (fci / fyi)=k1*Ti+k2*Ti 2 +k3*Ti 3 .

[0055] The node temperature-corrected normalized frequency difference obtaining submodule is configured to obtain node temperature-corrected normalized frequency difference fc according to frequency difference measurement value fci and temperature measurement value Ti, master node frequency measurement value fyi, and node temperature correction coefficient K, and according to fc=[fci-fyi*(1+k1*Ti+k2*Ti2+k3*Ti3)] / fyi.

[0056] The node normalized frequency difference mean value obtaining submodule is configured to obtain node normalized frequency difference mean value fm by averaging multiple node temperature-corrected normalized frequency difference fc measurement results of the node within a period of time.

[0057] The node normalized frequency difference standard deviation obtaining submodule is configured to obtain node normalized frequency difference standard deviation fs by calculating standard deviation of multiple node temperature-corrected normalized frequency difference fc measurement results of the node within a period of time.

[0058] The node frequency deviation aging rate obtaining submodule is configured to obtain node frequency deviation aging rate λ by fitting node temperature-corrected normalized frequency difference mean value fmi within a period of time according to fmi=λ*T+c.

[0059] a node time difference mean value submodule configured to average a plurality of time difference measurement results of the node in a period of time to obtain a node time difference mean value ΔTm;

[0060] a node time difference standard deviation submodule configured to calculate a standard deviation of a plurality of time difference measurement results of the node in a period of time to obtain a node time difference standard deviation ΔTs;

[0061] a superior node frequency deviation submodule configured to subtract a normalized frequency deviation of a superior node from the normalized frequency deviation of the node to obtain a frequency deviation between the node and the superior node fuc;

[0062] a superior node time deviation submodule configured to subtract a time deviation of a superior node from the time deviation of the node to obtain a time deviation between the node and the superior node ΔTuc;

[0063] a subordinate node frequency deviation submodule configured to subtract a normalized frequency deviation of a subordinate node from the normalized frequency deviation of the node to obtain a frequency deviation between the node and the subordinate node fnc;

[0064] a subordinate node time deviation submodule configured to subtract a time deviation of a subordinate node from the time deviation of the node to obtain a time deviation between the node and the subordinate node ΔTnc;

[0065] a node network delay mean value submodule configured to average network delay measurement values of the node in a period of time to obtain a node network delay mean value ΔTnm;

[0066] a node network delay standard deviation submodule configured to calculate a standard deviation of network delay measurement values of the node in a period of time to obtain a node network delay standard deviation ΔTns.

[0067] Optionally, the real-time measurement information division module comprises:

[0068] an abnormal type division submodule configured to determine an abnormal node range by analyzing the time-frequency node characteristic value, and divide the time-frequency node abnormal type into a single-node time-frequency synchronization data packet loss fault, a single-node synchronization network instability fault, a single-node time-frequency performance attenuation threshold fault, a single-node time-frequency performance mutation fault, a regional time-frequency synchronization data packet loss fault, and a regional synchronization network delay instability fault by analyzing the single-node time-frequency parameter evolution law and the cluster-internal neighboring node time-frequency parameter single-node synchronization network delay instability fault evolution law.

[0069] The single-node time-frequency synchronization data packet loss fault refers to that the node monitoring data can be collected, but the node time is not updated, the time-frequency characteristics of the subordinate node are consistent with those of the node, and the time-frequency characteristics of the superior node are normal.

[0070] The single-node synchronization network delay instability fault refers to that the node network delay standard deviation exceeds a threshold value.

[0071] Single node synchronization network instability fault refers to the node existing network offline state;

[0072] Single node time-frequency performance attenuation threshold fault refers to the node frequency deviation aging rate exceeding the threshold, while the superior node time-frequency characteristics are normal;

[0073] Single node time-frequency performance mutation fault refers to the comprehensive evaluation of the node normalized frequency difference mean, the node normalized frequency difference standard deviation, the node time difference mean, and the node time difference standard deviation. If the evaluation result exceeds the threshold, while the superior node time-frequency characteristics are normal, and the subordinate node time-frequency performance is consistent with the node, it is considered a fault.

[0074] Regional time-frequency synchronization data packet loss fault refers to the existence of data packet loss fault in multiple nodes within the area;

[0075] Regional synchronization network delay instability fault refers to the existence of network delay instability fault in multiple nodes within the area;

[0076] Regional synchronization network instability fault refers to the existence of network offline state in multiple nodes within the area;

[0077] Time-frequency device batch quality fault refers to the existence of performance attenuation threshold fault or time-frequency performance mutation fault in multiple devices in the running device.

[0078] Optionally, the evaluation verification set performance module includes:

[0079] The pruning submodule of the decision tree is used to evaluate the performance of the decision tree in the verification set according to the precision and recall rate. If the performance cannot reach the set threshold, the pruning strategy is defined based on the hyperparameter probability distribution. The Markov chain Monte Carlo chain is initialized, and the feature quantity of the real-time measurement information of the time-frequency node is considered to ensure that the pruning strategy is optimized for the actual application scenario.

[0080] The hyperparameter probability distribution updating submodule is used to sample the pruning strategy and evaluate the theoretical performance of each pruning strategy on the verification set. Based on the evaluation performance, the hyperparameter probability distribution is updated to gradually optimize the pruning strategy.

[0081] The retraining decision tree submodule is used to retrain the pruned decision tree with the updated hyperparameters, train the decision tree on the training set, and evaluate its performance on the verification set. The above process is repeated until the performance of the decision tree reaches the set threshold or the number of iterations exceeds the preset upper limit.

[0082] Therefore, the cause of the communication network anomaly in the station area can be quickly diagnosed, and the abnormal diagnosis result is given through the problem characteristic value, and the diagnosis effect is remarkable in the pilot station area deploying the algorithm. The accuracy of the intelligent diagnosis of the scheme can be improved according to the time of accumulating problem characteristics, and the detection rate and hit rate of the diagnosis of the communication network anomaly in the station area can be improved through continuous training of the model for new problem data sets, so that the time cost and economic cost of the original artificial troubleshooting and on-site problem positioning are greatly reduced. BRIEF DESCRIPTION OF DRAWINGS

[0083] The exemplary embodiments of this application can be more fully understood with reference to the following drawings:

[0084] Figure 1 A flowchart of a node anomaly diagnosis method based on a carrier time-frequency synchronization network cluster according to the embodiment;

[0085] Figure 2 A schematic diagram of establishing a topological relationship of each terminal node according to the embodiment;

[0086] Figure 3 A schematic diagram of a node anomaly diagnosis system based on a carrier time-frequency synchronization network cluster according to the embodiment. DETAILED DESCRIPTION

[0087] Exemplary embodiments of the present application will now be described with reference to the accompanying drawings; however, the present application can be implemented in many different forms and is not limited to the embodiments described herein, which are provided to fully and completely disclose the present application and to fully convey the scope of the present application to those skilled in the art. The terms used in the exemplary embodiments shown in the drawings are not limited to the present application. In the drawings, the same elements / elements are denoted by the same reference numerals.

[0088] Unless otherwise defined, the terms used herein (including technical terms) have meanings commonly understood by those skilled in the art. In addition, it is to be understood that the terms defined in commonly used dictionaries should be interpreted as having a meaning that is consistent with their meaning in the context of the relevant art and not be interpreted in an idealized or overly formal sense.

[0089] According to a first aspect of the present application, a node anomaly diagnosis method 100 based on a carrier time-frequency synchronization network cluster is provided, as shown in Figure 1 The method 100 comprises:

[0090] S101: determining a carrier-based station area time-frequency synchronization network, and collecting real-time measurement information of time-frequency nodes based on the carrier time-frequency network;

[0091] S102: Divide the time-frequency node real-time measurement information into feature values and abnormal types, extract the feature values of the time-frequency node real-time measurement information, obtain time-frequency node feature values, determine the abnormal node range by analyzing the time-frequency node feature values, and divide the time-frequency node abnormal types by analyzing the single node time-frequency parameter evolution law and the cluster-internal neighboring node time-frequency parameter evolution law;

[0092] S103: Randomly select a part of data from the time-frequency node real-time measurement information as a validation set, and the remaining data as a training set, learn a decision tree from the training set based on the decision tree generation algorithm C4.5;

[0093] S104: Evaluate the performance of the decision tree in the validation set according to the precision and recall rate, if the performance cannot reach the set threshold, prune the decision tree, retrain the pruned decision tree, until the performance of the decision tree reaches the set threshold or the iteration number exceeds the preset upper limit.

[0094] Specifically, the carrier-based district area time-frequency synchronization network includes a carrier master clock node and a plurality of carrier slave clock nodes, and the topology relationship between each terminal node is established based on the power carrier network, and then a synchronization system is established in the entire network, as shown in the reference Figure 2 In the topology: T-GM is the network master node, which is the master clock; T-TC is a relay node, which is both a slave clock and a transparent clock; T-SC is the end node, which is a slave clock.

[0095] T-GM synchronizes the time and frequency of other nodes in the network through the carrier, and simultaneously collects the time-frequency parameter real-time measurement information of each node.

[0096] The specific analysis process and method are as follows:

[0097] I) Collect the real-time measurement results of each node in the district area, and the specific data items are shown in Table 1.

[0098] Table 1 Time-frequency node real-time measurement data items

[0099] Acquisition item number Acquisition item name 1 Master device number 2 Superior device number 3 Subordinate device number 4 Frequency error 5 Time error 6 Network delay 7 Last time-frequency synchronization time 8 Ambient temperature

[0100] II) Based on the real-time measurement results, the feature values in Table 2 are extracted, and these feature values are used as the input of the abnormal diagnosis algorithm.

[0101] Table 2 Time-frequency node feature values

[0102] Feature number Feature name 1 Frequency temperature correction coefficient 2 Normalized frequency difference after node temperature correction 3 Node normalized frequency difference mean 4 Node normalized frequency difference standard deviation 5 Node frequency deviation aging rate 6 Node time difference mean 7 Node time difference standard deviation 8 Frequency deviation between superior node 9 Time deviation between superior node 10 Frequency deviation between subordinate node 11 Time deviation between subordinate node 12 Node network delay mean 13 Node network delay standard deviation

[0103] Frequency temperature correction coefficient K: According to the accurate frequency value fyi of the master clock, the frequency difference measurement value fci of the clock, the real-time temperature measurement value Ti, the fitting is carried out according to (fci / fyi)=k1*Ti+k2*Ti2+k3*Ti3, and the frequency correction coefficient K (k1, k2, k3) is obtained online.

[0104] Node temperature corrected normalized frequency difference fc: Based on the frequency difference measurement value fci and the temperature measurement value Ti, the master node frequency measurement value fyi, and the node temperature correction coefficient K, according to fc=[fci-fyi*(1+k1*Ti+k2*Ti2+k3*Ti3)] / fyi.

[0105] Node normalized frequency difference average fm: The average of multiple fc measurement results of the node in a period of time is obtained.

[0106] Node normalized frequency difference standard deviation fs: The standard deviation of multiple fc measurement results of the node in a period of time is obtained.

[0107] Node frequency deviation aging rate λ: The temperature corrected normalized frequency difference average fmi of the node in a period of time is fitted according to fmi=λ*T+c, and λ is obtained.

[0108] Node time difference average ΔTm: The average of multiple time difference measurement results of the node in a period of time is obtained.

[0109] Node time difference standard deviation ΔTs: The standard deviation of multiple time difference measurement results of the node in a period of time is obtained.

[0110] Frequency deviation fuc between the upper node: The normalized frequency difference of the node is subtracted from the normalized frequency difference of the upper node at the same time to obtain fuc.

[0111] Time deviation ΔTuc between the upper node: The time deviation of the node is subtracted from the time deviation of the upper node at the same time to obtain ΔTuc.

[0112] Frequency deviation fnc between the lower node: The normalized frequency difference of the node is subtracted from the normalized frequency difference of the lower node at the same time to obtain fuc.

[0113] Time deviation ΔTnc between the lower node: The time deviation of the node is subtracted from the time deviation of the lower node at the same time to obtain ΔTuc.

[0114] Node network delay average ΔTnm: The network delay measurement value of the node in a period of time is averaged to obtain ΔTnm.

[0115] Node network delay standard deviation ΔTns: The network delay measurement value of the node in a period of time is calculated to obtain ΔTns.

[0116] III) Based on the time-frequency network, the abnormal situation is divided into types, as shown in Table 3.

[0117] Table 3 Abnormal type division

[0118]

[0119]

[0120] Single node time-frequency synchronization data packet loss fault: node monitoring data can be collected, but the node time is not updated at one time, and the time-frequency characteristics of the lower node are consistent with the node, while the time-frequency characteristics of the upper node are normal.

[0121] Single node synchronization network delay instability fault: the standard deviation of node network delay exceeds the threshold.

[0122] Single node synchronization network instability fault: the node has network offline state.

[0123] Single node time-frequency performance decay threshold fault: the node frequency deviation aging rate exceeds the threshold, and the time-frequency characteristics of the upper node are normal.

[0124] Single node time-frequency performance mutation fault: the evaluation result is evaluated by the node normalized frequency difference mean, the node normalized frequency difference standard deviation, the node time difference mean and the node time difference standard deviation, and the evaluation result exceeds the threshold, and the time-frequency characteristics of the upper node are normal, and the time-frequency performance of the lower node is consistent with the node.

[0125] Regional time-frequency synchronization data packet loss fault: multiple nodes in the area exist data packet loss fault.

[0126] Regional synchronization network delay instability fault: multiple nodes in the area exist network delay instability fault.

[0127] Regional synchronization network instability fault: multiple nodes in the area exist network offline state.

[0128] Time-frequency equipment batch quality fault: multiple devices in the running device exist performance decay threshold fault or time-frequency performance mutation fault.

[0129] IV) Randomly select a part of the data as the verification set, and the rest as the training set, and use the decision tree generation algorithm C4.5 to learn a decision tree from the training data.

[0130] Based on the real-time measurement results of each node in the collection area, they are divided into characteristic quantities and abnormal types for analysis. First, a part of the data is randomly selected as the validation set, and the rest is used as the training set. A decision tree is learned from the training data using the decision tree generation algorithm C4.5. The characteristic quantities are shown in Table 2, including frequency temperature correction coefficient, node temperature correction normalized frequency difference, etc.; the abnormal types are shown in Table 3, including single node time-frequency synchronization data packet loss fault, regional synchronization network delay instability fault, etc.

[0131] V) The performance of the decision tree on the validation set is evaluated using precision and recall. If the performance does not meet the threshold, the decision tree is pruned. A hyperparameter probability distribution is defined for pruning and the Markov chain Monte Carlo chain is initialized.

[0132] The performance of the decision tree on the validation set is evaluated using precision and recall. If the performance does not meet the threshold, the decision tree needs to be pruned and optimized. A hyperparameter probability distribution is defined for pruning, and the Markov chain Monte Carlo chain is initialized to improve the classification effect of the decision tree. During the pruning process, the characteristic quantities in Table 2 are mainly considered, such as node normalized frequency difference mean, node time difference standard deviation, etc., to ensure that the pruning strategy is optimized for the actual application scenario.

[0133] VI) The pruning strategy is sampled using this Markov chain Monte Carlo chain, and the theoretical performance of each strategy on the validation set is calculated. The evaluation results are used to update the hyperparameter probability distribution using the Bayes formula.

[0134] The pruning strategy is sampled using the Markov chain Monte Carlo method, and the theoretical performance of each strategy on the validation set is calculated. Based on the evaluation results, the hyperparameter probability distribution is updated using the Bayes formula, so that the pruning strategy is gradually optimized, improving the robustness and generalization ability of the model. This process realizes the dynamic adjustment of the pruning strategy through comprehensive analysis of the characteristic quantities, ensuring the adaptability of the model in complex time-frequency network environments.

[0135] VII) The pruning decision tree with updated hyperparameter values is used to train the decision tree on the training set, and the performance is evaluated on the validation set. Repeat the above process VI) until the performance reaches the threshold or the iteration exceeds the set number of times.

[0136] The pruning decision tree with updated hyperparameters is retrained, and the decision tree is trained on the training set, and its performance is evaluated on the validation set. Repeat the above process until the performance of the decision tree reaches the set threshold or the number of iterations exceeds the preset upper limit. This loop process guarantees the accuracy and stability of the decision tree in time-frequency node anomaly diagnosis, and improves the accuracy and reliability of anomaly detection through the optimized pruning strategy.

[0137] Therefore, the cause of the communication network anomaly in the transformer area can be quickly diagnosed, and the abnormal diagnosis result is given through the problem characteristic value. In the pilot transformer area deploying the algorithm, the diagnosis effect is remarkable. The accuracy of the intelligent diagnosis of the scheme can be improved according to the time of accumulating problem characteristics, and the detection rate and hit rate of the diagnosis of the communication network anomaly in the transformer area can be improved through continuous training of the model for new problem data sets, thereby greatly reducing the time cost and economic cost of original artificial troubleshooting and on-site problem positioning.

[0138] Optionally, a carrier-based transformer area time-frequency synchronization network is determined, and time-frequency node real-time measurement information is collected based on the carrier time-frequency network, including:

[0139] The transformer area time-frequency synchronization network includes a plurality of terminal nodes, that is, one carrier master clock node and a plurality of carrier slave clock nodes;

[0140] Based on the power carrier network, a topology relationship of each terminal node is established, and a synchronization system is established in the entire power carrier network. In the topology relationship: the T-GM is a network master node and a master clock; the T-TC is a relay node and is both a slave clock and a transparent clock; and the T-SC is a terminal node and is a slave clock.

[0141] The carrier is collected through the T-GM master node, the time-frequency synchronization of other nodes in the network is performed, and time-frequency node real-time measurement information is collected, including: the device number, the superior device number, the subordinate device number, the frequency error, the time error, the network delay, the last time-frequency synchronization time, and the environmental temperature.

[0142] Optionally, the characteristic quantity of the time-frequency node real-time measurement information is extracted, and a time-frequency node characteristic value is obtained, including:

[0143] The time-frequency node characteristic value includes: a frequency temperature correction coefficient, a node temperature corrected normalized frequency difference, a node normalized frequency difference mean value, a node normalized frequency difference standard deviation, a node frequency deviation aging rate, a node time difference mean value, a node time difference standard deviation, a frequency deviation between the superior node, a time deviation between the superior node, a frequency deviation between the subordinate node, a time deviation between the superior node, a node network delay mean value, and a node network delay standard deviation.

[0144] According to the accurate frequency value fyi of the master clock, the frequency difference measurement value fci of the slave clock, and the real-time temperature measurement value Ti, the fitting is performed according to (fci / fyi)=k1*Ti+k2*Ti 2 +k3*Ti 3 , and the frequency temperature correction coefficient K (k1, k2, k3) is obtained online.

[0145] Based on the frequency difference measurement value fci and the temperature measurement value Ti, the master node frequency measurement value fyi, and the node temperature correction coefficient K, the node temperature corrected normalized frequency difference fc is obtained according to fc=[fci-fyi*(1+k1*Ti+k2*Ti2+k3*Ti3)] / fyi;

[0146] The node normalized frequency difference average value fm is obtained by averaging the node temperature corrected normalized frequency difference fc measurement results of the node within a period of time;

[0147] The node normalized frequency difference standard deviation fs is obtained by calculating the standard deviation of the node temperature corrected normalized frequency difference fc measurement results of the node within a period of time;

[0148] The node frequency deviation aging rate λ is obtained by fitting the node temperature corrected normalized frequency difference average value fmi according to fmi=λ*T+c within a period of time;

[0149] The node time difference average value ΔTm is obtained by averaging the time difference measurement results of the node within a period of time;

[0150] The node time difference standard deviation ΔTs is obtained by calculating the standard deviation of the time difference measurement results of the node within a period of time;

[0151] The node normalized frequency difference is subtracted from the normalized frequency difference of the same period upper node to obtain the frequency deviation fuc between the upper node;

[0152] The node time deviation is subtracted from the time deviation of the same period upper node to obtain the time deviation ΔTuc between the upper node;

[0153] The node normalized frequency difference is subtracted from the normalized frequency difference of the same period lower node to obtain the frequency deviation fnc between the lower node;

[0154] The node time deviation is subtracted from the time deviation of the same period lower node to obtain the time deviation ΔTnc between the lower node;

[0155] The network delay measurement value of the node within a period of time is averaged to obtain the node network delay average value ΔTnm;

[0156] The standard deviation of the network delay measurement value of the node within a period of time is calculated to obtain the node network delay standard deviation ΔTns.

[0157] Optionally, the abnormal node range is determined by analyzing the time-frequency node characteristic value, and the time-frequency node abnormal type is divided by analyzing the single node time-frequency parameter evolution law and the cluster adjacent node time-frequency parameter evolution law, including:

[0158] The abnormal node range is determined by analyzing the time-frequency node characteristic value, the time-frequency node abnormal type is divided into a single node time-frequency synchronization data packet loss fault, a single node synchronization network networking unstable fault, a single node time-frequency performance decay threshold fault, a single node time-frequency performance mutation fault, a regional time-frequency synchronization data packet loss fault, and a regional synchronization network delay unstable fault by analyzing the single node time-frequency parameter evolution law and the cluster adjacent node time-frequency parameter single node synchronization network delay unstable fault evolution law.

[0159] The single node time-frequency synchronization data packet loss fault refers to that the node monitoring data can be collected, but the node time is not updated once, the lower node time-frequency characteristics are consistent with the node, and the upper node time-frequency characteristics are normal.

[0160] The single node synchronization network delay unstable fault refers to that the node network delay standard deviation exceeds the threshold.

[0161] The single node synchronization network networking unstable fault refers to that the node has a network offline state.

[0162] The single node time-frequency performance decay threshold fault refers to that the node frequency deviation aging rate exceeds the threshold, and the upper node time-frequency characteristics are normal.

[0163] The single node time-frequency performance mutation fault refers to that the evaluation result exceeds the threshold through comprehensive evaluation of the node normalized frequency difference mean, the node normalized frequency difference standard deviation, the node time difference mean and the node time difference standard deviation, the upper node time-frequency characteristics are normal, and the lower node time-frequency performance is consistent with the node.

[0164] The regional time-frequency synchronization data packet loss fault refers to that multiple nodes in the area exist data packet loss faults.

[0165] The regional synchronization network delay unstable fault refers to that multiple nodes in the area exist network delay unstable faults.

[0166] The regional synchronization network networking unstable fault refers to that multiple nodes in the area exist network offline states.

[0167] The time-frequency equipment batch quality fault refers to that multiple devices in the running devices exist performance decay threshold faults or time-frequency performance mutation faults.

[0168] Optionally, the performance of the decision tree in the verification set is evaluated according to the precision and recall rate, if the performance cannot reach the set threshold, the decision tree is pruned, and the pruned decision tree is retrained until the performance of the decision tree reaches the set threshold or the iteration number exceeds the preset upper limit, comprising:

[0169] According to the precision and recall rate, the performance of the decision tree in the verification set is evaluated, if the performance cannot reach the set threshold, the hyperparameter probability distribution is defined to prune the decision tree, the Markov chain Monte Carlo chain is initialized, the feature quantity of the real-time measurement information of the time-frequency node is considered, and it is ensured that the pruning strategy is optimized for the actual application scene.

[0170] The pruning strategy is sampled, and the theoretical performance of each pruning strategy on the verification set is evaluated, and the hyperparameter probability distribution is updated based on the evaluation performance, so that the pruning strategy is gradually optimized.

[0171] The pruned decision tree is retrained by the updated hyperparameters, and the decision tree is trained on the training set, and its performance is evaluated on the verification set, and the above process is repeated until the performance of the decision tree reaches the set threshold or the iteration number exceeds the preset upper limit.

[0172] Therefore, the cause of the communication network anomaly in the station area can be quickly diagnosed, and the abnormal diagnosis result is given through the problem characteristic value, and the diagnosis effect is remarkable in the pilot station area deploying the algorithm. The accuracy of intelligent diagnosis of the scheme can be improved according to the accumulated problem characteristic time, and the detection rate and hit rate of the diagnosis of the communication network anomaly in the station area can be improved through continuous training of the model for new problem data sets, which greatly reduces the time cost and economic cost of original artificial troubleshooting and on-site problem positioning.

[0173] According to another aspect of the present application, a node anomaly diagnosis system 300 based on a carrier time-frequency synchronization network cluster is also provided, as shown in Figure 3 The system 300 comprises:

[0174] A real-time measurement information determination module 310 is configured to determine a carrier-based station area time-frequency synchronization network, and collect real-time measurement information of time-frequency nodes based on the carrier time-frequency network;

[0175] A real-time measurement information division module 320 is configured to divide the real-time measurement information of the time-frequency nodes into feature quantities and anomaly types, extract the feature quantities of the real-time measurement information of the time-frequency nodes, obtain time-frequency node characteristic values, determine the abnormal node range by analyzing the time-frequency node characteristic values, and divide the time-frequency node anomaly types by analyzing the single node time-frequency parameter evolution law and the cluster adjacent node time-frequency parameter evolution law;

[0176] A decision tree learning module 330 is configured to randomly select a part of data from the real-time measurement information of the time-frequency nodes as a verification set, and the remaining part of data as a training set, and learn a decision tree from the training set based on a decision tree generation algorithm C4.5;

[0177] The evaluation verification set performance module 340 is configured to evaluate the performance of the decision tree in the verification set according to the precision and recall rate, prune the decision tree if the performance cannot reach the set threshold, retrain the pruned decision tree, and repeat the process until the performance of the decision tree reaches the set threshold or the number of iterations exceeds the preset upper limit.

[0178] Optionally, the real-time measurement information determining module comprises:

[0179] The time-frequency synchronization network determining submodule is configured to determine that the time-frequency synchronization network comprises a plurality of terminal nodes, i.e., one carrier master clock node and a plurality of carrier slave clock nodes.

[0180] The topology relationship establishing submodule is configured to establish a topology relationship among the terminal nodes based on the power carrier network, and further establish a synchronization system in the entire power carrier network, wherein in the topology relationship, the T-GM is a network master node and a master clock; the T-TC is a relay node, which is both a slave clock and a transparent clock; and the T-SC is an end node, which is a slave clock.

[0181] The time-frequency node real-time measurement information collecting submodule is configured to collect time-frequency node real-time measurement information by the T-GM master node through the carrier, synchronize time and frequency among other nodes in the network, and collect the time-frequency node real-time measurement information, which comprises a device number, a superior device number, an inferior device number, a frequency error, a time error, a network delay, a last time-frequency synchronization time, and an ambient temperature.

[0182] Optionally, the real-time measurement information dividing module comprises:

[0183] The time-frequency node characteristic value comprises a frequency temperature correction coefficient, a node temperature-corrected normalized frequency difference, a node normalized frequency difference mean value, a node normalized frequency difference standard deviation, a node frequency deviation aging rate, a node time difference mean value, a node time difference standard deviation, a frequency deviation between the node and a superior node, a time deviation between the node and the superior node, a frequency deviation between the node and an inferior node, a time deviation between the node and the inferior node, a node network delay mean value, and a node network delay standard deviation.

[0184] The frequency temperature correction coefficient obtaining submodule is configured to perform fitting according to an accurate frequency value fyi of the master clock, a frequency difference measurement value fci of the slave clock, and a real-time temperature measurement value Ti, according to (fci / fyi)=k1*Ti+k2*Ti+k3*Ti, to obtain the frequency temperature correction coefficient K (k1, k2, k3) online. 2 3

[0185] ​​A node temperature correction after normalization frequency difference submodule is configured to obtain a node temperature correction after normalization frequency difference fc based on a frequency difference measurement value fci, a temperature measurement value Ti, a master node frequency measurement value fyi, and a node temperature correction coefficient K, according to fc = [fci-fyi*(1+k1*Ti+k2*Ti2+k3*Ti3)] / fyi;

[0186] A node normalization frequency difference average submodule is configured to obtain a node normalization frequency difference average fm by averaging a plurality of node temperature correction after normalization frequency difference fc measurement results of a node within a period of time;

[0187] A node normalization frequency difference standard deviation submodule is configured to obtain a node normalization frequency difference standard deviation fs by calculating a standard deviation of a plurality of node temperature correction after normalization frequency difference fc measurement results of a node within a period of time;

[0188] A node frequency deviation aging rate submodule is configured to obtain a node frequency deviation aging rate λ by fitting a node temperature correction after normalization frequency difference average fmi according to fmi = λ*T + c within a period of time;

[0189] A node time difference average submodule is configured to obtain a node time difference average ΔTm by averaging a plurality of time difference measurement results of a node within a period of time;

[0190] A node time difference standard submodule is configured to obtain a node time difference standard deviation ΔTs by calculating a standard deviation of a plurality of time difference measurement results of a node within a period of time;

[0191] A frequency deviation from a superior node submodule is configured to obtain a frequency deviation from a superior node fuc by subtracting a superior node normalization frequency difference from a node normalization frequency difference at the same time;

[0192] A time deviation from a superior node submodule is configured to obtain a time deviation from a superior node ΔTuc by subtracting a superior node time deviation from a node time deviation at the same time;

[0193] A frequency deviation from an inferior node submodule is configured to obtain a frequency deviation from an inferior node fnc by subtracting an inferior node normalization frequency difference from a node normalization frequency difference at the same time;

[0194] A time deviation from an inferior node submodule is configured to obtain a time deviation from an inferior node ΔTnc by subtracting an inferior node time deviation from a node time deviation at the same time;

[0195] A node network delay average submodule is configured to obtain a node network delay average ΔTnm by averaging network delay measurement values of a node within a period of time;

[0196] The node network delay standard deviation submodule is configured to calculate the standard deviation of the network delay measurement values of the node within a period of time to obtain a node network delay standard deviation ΔTns.

[0197] Optionally, the real-time measurement information division module comprises:

[0198] The abnormal type division submodule is configured to determine an abnormal node range by analyzing the time-frequency node characteristic value, and divide the time-frequency node abnormal type into a single-node time-frequency synchronization data packet loss fault, a single-node synchronization network unstable fault, a single-node time-frequency performance attenuation threshold fault, a single-node time-frequency performance mutation fault, a regional time-frequency synchronization data packet loss fault, and a regional synchronization network delay unstable fault by analyzing the single-node time-frequency parameter evolution law and the single-node synchronization network delay unstable fault evolution law of the neighboring nodes in the cluster.

[0199] The single-node time-frequency synchronization data packet loss fault refers to that the node monitoring data can be collected, but the node time is not updated at last time, the time-frequency characteristics of the lower-level nodes are consistent with those of the node, and the time-frequency characteristics of the upper-level node are normal.

[0200] The single-node synchronization network delay unstable fault refers to that the node network delay standard deviation exceeds a threshold value.

[0201] The single-node synchronization network unstable fault refers to that the node has a network offline state.

[0202] The single-node time-frequency performance attenuation threshold fault refers to that the node frequency deviation aging rate exceeds a threshold value, and the time-frequency characteristics of the upper-level node are normal.

[0203] The single-node time-frequency performance mutation fault refers to that the evaluation result exceeds a threshold value through comprehensive evaluation of the node normalized frequency difference mean value, the node normalized frequency difference standard deviation, the node time difference mean value, and the node time difference standard deviation, and the time-frequency performance of the lower-level nodes is consistent with that of the node, and the time-frequency characteristics of the upper-level node are normal.

[0204] The regional time-frequency synchronization data packet loss fault refers to that multiple nodes in the area exist data packet loss faults.

[0205] The regional synchronization network delay unstable fault refers to that multiple nodes in the area exist network delay unstable faults.

[0206] The regional synchronization network unstable fault refers to that multiple nodes in the area exist network offline states.

[0207] The time-frequency device batch quality fault refers to that multiple devices in the running devices exist performance attenuation threshold faults or time-frequency performance mutation faults.

[0208] Optionally, the evaluation verification set performance module comprises:

[0209] The pruning submodule of the decision tree is used for evaluating the performance of the decision tree in the verification set according to the precision and recall rate, if the performance cannot reach the set threshold, the pruning strategy is defined by the hyperparameter probability distribution, the Markov chain Monte Carlo chain is initialized, the feature quantity of the real-time measurement information of the time-frequency node is considered, and it is ensured that the pruning strategy is optimized for the actual application scene.

[0210] The updating submodule of the hyperparameter probability distribution is used for sampling the pruning strategy, evaluating the theoretical performance of each pruning strategy on the verification set, updating the hyperparameter probability distribution based on the evaluation performance, and gradually optimizing the pruning strategy.

[0211] The retraining submodule of the decision tree is used for retraining the pruned decision tree by the updated hyperparameter, training the decision tree on the training set, evaluating the performance of the decision tree on the verification set, and repeating the above process until the performance of the decision tree reaches the set threshold or the iteration number exceeds the preset upper limit.

[0212] The node anomaly diagnosis system 300 in the carrier time-frequency synchronization network cluster based on an embodiment of the application corresponds to the node anomaly diagnosis method 100 in the carrier time-frequency synchronization network cluster based on another embodiment of the application, which will not be described here.

[0213] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can 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 code. The solutions in the embodiments of the present application can be implemented in various computer languages, such as object-oriented programming language Java and interpreted scripting language JavaScript.

[0214] The present application is described with reference to flowcharts and / or block diagrams according to the methods, devices (systems), and computer program products of the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of the flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to the 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 produce a device that implements the functions specified in the flowcharts and / or block diagrams. Figure 1 The device that implements the functions specified in one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0215] These computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions Figure 1 one or more functions specified in the flow Figure 1 one or more blocks or multiple blocks.

[0216] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions Figure 1 one or more functions specified in the flow Figure 1 one or more blocks or multiple blocks.

[0217] While the preferred embodiments of the application have been described, additional variations and modifications can be made to the embodiments by those skilled in the art once they learn of the basic inventive concepts. Therefore, the appended claims are intended to encompass within their scope all such variations and modifications as are included within the scope of the application.

[0218] Obviously, numerous modifications and variations of the present application are possible in light of the above teachings. It is therefore to be understood that within the scope of the appended claims and their equivalents, the application can be practiced otherwise than as specifically described.

Claims

1. A method for diagnosing node anomalies in a carrier-based time-frequency synchronization network cluster, characterized in that, include: Establish a carrier-based time-frequency synchronization network for the distribution area, and collect real-time measurement information of the time-frequency nodes based on the carrier-based time-frequency network; The real-time measurement information of the time-frequency nodes is divided into feature quantities and anomaly types. The feature quantities of the real-time measurement information of the time-frequency nodes are extracted to obtain the feature values ​​of the time-frequency nodes. The range of abnormal nodes is determined by analyzing the feature values ​​of the time-frequency nodes. The anomaly types of time-frequency nodes are classified by analyzing the evolution law of time-frequency parameters of single nodes and the evolution law of time-frequency parameters of neighboring nodes in the cluster. The time-frequency node characteristic values ​​include: frequency temperature correction coefficient, normalized frequency difference after node temperature correction, mean of normalized frequency difference, standard deviation of normalized frequency difference, node frequency deviation aging rate, mean of node time difference, standard deviation of node time difference, frequency deviation with upper-level node, time deviation with upper-level node, frequency deviation with lower-level node, time deviation with upper-level node, mean of node network delay, and standard deviation of node network delay. A portion of the data is randomly selected from the real-time measurement information of the time-frequency nodes as a validation set, and the remaining data is used as a training set. A decision tree is learned from the training set based on the decision tree generation algorithm C4.

5. The performance of the decision tree on the validation set is evaluated based on precision and recall. If the performance cannot reach the set threshold, a hyperparameter probability distribution is defined to prune the decision tree. The Markov chain Monte Carlo is initialized, and the feature quantity of real-time measurement information of time and frequency nodes is considered to ensure that the pruning strategy is optimized for the actual application scenario. The pruning strategies are sampled, and the theoretical performance of each pruning strategy on the validation set is evaluated. The hyperparameter probability distribution is updated based on the evaluation performance, so that the pruning strategies are gradually optimized. The pruned decision tree is retrained using the updated hyperparameters, and its performance is evaluated on the validation set. This process is repeated until the performance of the decision tree reaches a set threshold or the number of iterations exceeds a preset limit. The time-frequency node feature values ​​are used as input to the anomaly diagnosis algorithm to diagnose the cause of communication network anomalies within the transformer area, and the anomaly diagnosis results are given based on the problematic feature values.

2. The method according to claim 1, characterized in that, Determine the carrier-based time-frequency synchronization network for the distribution area, and collect real-time measurement information of the time-frequency nodes based on the carrier-based time-frequency network, including: The time-frequency synchronization network of the distribution area is determined to include multiple terminal nodes, namely one carrier master clock node and multiple carrier slave clock nodes; Based on the power line carrier network, a topology relationship is established between each terminal node, thereby establishing a synchronization system in the entire power line carrier network. In this topology relationship: T-GM is the network master node and is the master clock; T-TC is the relay node, which is both a slave clock and a transparent clock; T-SC is the terminal node and is a slave clock. The T-GM master node carrier performs time-frequency synchronization with other nodes in the network, and simultaneously collects real-time measurement information of the time-frequency nodes. The real-time measurement information of the time-frequency nodes includes: this device number, the number of the superior device, the number of the subordinate device, the frequency error, the time error, the network delay, the last time-frequency synchronization time, and the ambient temperature.

3. The method according to claim 2, characterized in that, Extracting the feature quantities of the real-time measurement information of the time-frequency nodes to obtain the feature values ​​of the time-frequency nodes includes: Based on the accurate frequency value fyi of the master clock, the frequency difference measurement fci of the clock, and the real-time temperature measurement Ti, the formula is (fci / fyi) = k1*Ti + k2*Ti. 2 +k3*Ti 3 The frequency temperature correction coefficient K(k1,k2,k3) is obtained online by fitting the data. Based on the frequency difference measurement value fci and the temperature measurement value Ti, the master node frequency measurement value fyi, and the node temperature correction coefficient K, according to fc = [fci - fyi * (1 + k1 * Ti + k2 * Ti)] 2 +k3*Ti 3 )] / fyi, obtain the normalized frequency difference fc after node temperature correction; The mean normalized frequency difference fm of the node is obtained by averaging the normalized frequency difference fc measurement results of the node after multiple node temperature corrections over a period of time. The standard deviation of the normalized frequency difference fc is obtained by calculating the standard deviation of the node after multiple node temperature corrections over a period of time. The normalized mean frequency difference fm of the node after temperature correction over a period of time i According to fm i The node frequency deviation aging rate λ is obtained by fitting λ*T+c. The mean time difference ΔTm of a node is obtained by averaging the results of multiple time difference measurements of the node over a period of time. The standard deviation of the node's time difference, ΔTs, is obtained by calculating the standard deviation of multiple time difference measurements of the node over a period of time. Subtracting the normalized frequency difference of the parent node from the normalized frequency difference of the node yields the frequency deviation fuc between the parent and the parent node. Subtracting the time deviation of the parent node from the time deviation of the node at the same time gives the time deviation ΔTuc between the node and the parent node. Subtracting the normalized frequency difference of the lower-level nodes from the normalized frequency difference of the nodes at the same time yields the frequency deviation fnc between the nodes and the lower-level nodes. Subtracting the time deviation of the lower-level nodes in the same period from the node time deviation yields the time deviation ΔTnc between the node and the lower-level nodes. The average network latency measurement of a node over a period of time is used to obtain the mean network latency ΔTnm. The standard deviation of the network latency measurement of a node over a period of time is calculated to obtain the standard deviation of the node network latency ΔTns.

4. The method according to claim 1, characterized in that, The range of abnormal nodes is determined by analyzing the characteristic values ​​of the time-frequency nodes. The types of time-frequency node anomalies are classified by analyzing the evolution patterns of single-node time-frequency parameters and the evolution patterns of time-frequency parameters of neighboring nodes within the cluster. These include: The range of abnormal nodes is determined by analyzing the characteristic values ​​of the time-frequency nodes. By analyzing the evolution law of time-frequency parameters of a single node and the evolution law of time-frequency parameters of neighboring nodes in the cluster, the abnormal types of time-frequency nodes are classified as follows: single node time-frequency synchronization data packet loss fault, single node synchronization network delay instability fault, single node synchronization network networking instability fault, single node time-frequency performance degradation exceeding threshold fault, single node time-frequency performance sudden change fault, regional time-frequency synchronization data packet loss fault, regional synchronization network delay instability fault, regional synchronization network networking instability fault, and time-frequency equipment batch quality fault. Single-node time-frequency synchronization data packet loss fault refers to a situation where the node can collect monitoring data, but the node's last time was not updated, while the time-frequency characteristics of the lower-level node are consistent with this node, and the time-frequency characteristics of the upper-level node are normal. Single-node synchronous network latency instability fault refers to a node network latency standard deviation exceeding a threshold. A single-node synchronous network instability fault refers to a node being offline in the network. A single-node time-frequency performance degradation exceeding the threshold fault refers to a node frequency deviation aging rate exceeding the threshold, while the time-frequency characteristics of the upstream node are normal. A single-node time-frequency performance mutation fault refers to a comprehensive evaluation of the changes in the node's normalized frequency difference mean, node normalized frequency difference standard deviation, node time difference mean, and node time difference standard deviation. The evaluation result exceeds the threshold, while the time-frequency characteristics of the upper-level node are normal, and the time-frequency performance of the lower-level node is consistent with that node. Regional time-frequency synchronization data packet loss fault refers to data packet loss faults occurring at multiple nodes within the station area; Regional synchronization network latency instability fault refers to the network latency instability fault occurring at multiple nodes within the station area; The instability of the regional synchronization network indicates that multiple nodes within the station area are offline. Batch quality failure of time and frequency equipment refers to multiple devices in operation experiencing performance degradation exceeding the threshold or sudden changes in time and frequency performance.

5. A node anomaly diagnosis system in a carrier-based time-frequency synchronization network cluster, characterized in that, include: The real-time measurement information module is used to determine the carrier-based time-frequency synchronization network of the transformer area and to collect real-time measurement information of the time-frequency nodes based on the carrier-based time-frequency network. The real-time measurement information module is used to divide the real-time measurement information of the time-frequency nodes into feature quantities and anomaly types, extract the feature quantities of the real-time measurement information of the time-frequency nodes, obtain the feature values ​​of the time-frequency nodes, determine the range of abnormal nodes by analyzing the feature values ​​of the time-frequency nodes, and classify the anomaly types of the time-frequency nodes by analyzing the evolution law of the time-frequency parameters of a single node and the evolution law of the time-frequency parameters of neighboring nodes in the cluster. The time-frequency node characteristic values ​​include: frequency temperature correction coefficient, normalized frequency difference after node temperature correction, mean of normalized frequency difference, standard deviation of normalized frequency difference, node frequency deviation aging rate, mean of node time difference, standard deviation of node time difference, frequency deviation with upper-level node, time deviation with upper-level node, frequency deviation with lower-level node, time deviation with upper-level node, mean of node network delay, and standard deviation of node network delay. The decision tree learning module is used to randomly select a portion of the data from the real-time measurement information of the time-frequency nodes as a validation set and the remaining data as a training set, and learn a decision tree from the training set based on the decision tree generation algorithm C4.

5. The evaluation validation set performance module includes: The decision tree pruning submodule is used to evaluate the performance of the decision tree on the validation set based on precision and recall. If the performance cannot reach the set threshold, the hyperparameter probability distribution is defined to prune the decision tree. The Markov chain Monte Carlo chain is initialized, and the feature quantity of real-time measurement information of time and frequency nodes is considered to ensure that the pruning strategy is optimized for the actual application scenario. The hyperparameter probability distribution update submodule is used to sample the pruning strategies and evaluate the theoretical performance of each pruning strategy on the validation set. Based on the evaluation performance, the hyperparameter probability distribution is updated so that the pruning strategies are gradually optimized. The retraining decision tree submodule is used to retrain the pruned decision tree with updated hyperparameters, train the decision tree on the training set, evaluate its performance on the validation set, and repeat the above process until the performance of the decision tree reaches a set threshold or the number of iterations exceeds a preset upper limit. The time-frequency node feature values ​​are used as input to the anomaly diagnosis algorithm to diagnose the cause of communication network anomalies in the substation area, and the anomaly diagnosis results are given based on the problematic feature values.

6. The system according to claim 5, characterized in that, The real-time measurement information module is defined, including: The time-frequency synchronization network submodule is used to determine that the time-frequency synchronization network of the distribution area includes multiple terminal nodes, namely one carrier master clock node and multiple carrier slave clock nodes; The topology relationship establishment submodule is used to establish the topology relationship between each terminal node based on the power line carrier network, thereby establishing a synchronization system in the entire power line carrier network. In this topology relationship: T-GM is the network master node and is the master clock; T-TC is the relay node, which is both a slave clock and a transparent clock; T-SC is the terminal node and is a slave clock. The submodule for collecting real-time measurement information of time and frequency nodes is used to synchronize the time and frequency of other nodes in the network through the T-GM master node carrier, and at the same time collect the real-time measurement information of time and frequency nodes. The real-time measurement information of time and frequency nodes includes: this device number, the number of the superior device, the number of the subordinate device, the frequency error, the time error, the network delay, the last time and frequency synchronization time, and the ambient temperature.

7. The system according to claim 6, characterized in that, The real-time measurement information module is divided into several parts, including: The frequency-temperature correction coefficient submodule is used to calculate the frequency-temperature correction coefficient based on the accurate frequency value fyi of the master clock, the frequency difference measurement value fci of the clock, and the real-time temperature measurement value Ti, according to (fci / fyi) = k1*Ti + k2*Ti. 2 +k3*Ti 3 The frequency temperature correction coefficient K(k1,k2,k3) is obtained online by fitting the data. The normalized frequency difference submodule after obtaining node temperature correction is used to calculate the frequency difference based on the frequency difference measurement value fci, the temperature measurement value Ti, the master node frequency measurement value fyi, and the node temperature correction coefficient K, according to fc = [fci - fyi * (1 + k1 * Ti + k2 * Ti). 2 +k3*Ti 3 )] / fyi, obtain the normalized frequency difference fc after node temperature correction; The submodule for obtaining the normalized frequency difference mean of nodes is used to average the normalized frequency difference fc measurement results of nodes after multiple node temperature corrections over a period of time to obtain the normalized frequency difference mean fm of nodes. The node normalized frequency difference standard deviation submodule is used to calculate the standard deviation of the node normalized frequency difference fc measurement results after multiple node temperature corrections over a period of time, and obtain the node normalized frequency difference standard deviation fs. The node frequency deviation aging rate submodule is used to normalize the mean frequency deviation fm of the node after temperature correction over a period of time. i According to fm i The node frequency deviation aging rate λ is obtained by fitting λ*T+c. The submodule for obtaining the mean time difference of nodes is used to average the results of multiple time difference measurements of nodes over a period of time to obtain the mean time difference of nodes ΔTm. The node time difference standard submodule is used to calculate the standard deviation of the node time difference standard ΔTs by calculating the standard deviation of the node time difference standard from multiple time difference measurements over a period of time. The frequency deviation submodule between the node and the parent node is obtained, which is used to subtract the normalized frequency difference of the parent node from the normalized frequency difference of the node to obtain the frequency deviation fuc between the node and the parent node. The submodule for obtaining the time deviation from the parent node is used to subtract the time deviation of the parent node from the time deviation of the node to obtain the time deviation ΔTuc between the parent node and the parent node. The frequency deviation submodule between the node and the lower-level node is obtained, which is used to subtract the normalized frequency difference of the lower-level node from the normalized frequency difference of the node to obtain the frequency deviation fnc between the node and the lower-level node. The submodule for obtaining the time deviation with the lower-level node is used to subtract the time deviation of the lower-level node from the node's time deviation to obtain the time deviation ΔTnc with the lower-level node. The submodule for obtaining the average network latency of nodes is used to average the network latency measurements of nodes over a period of time to obtain the average network latency ΔTnm of the nodes. The submodule for obtaining the standard deviation of node network delay is used to calculate the standard deviation of the network delay measurement values ​​of the node over a period of time to obtain the standard deviation of node network delay ΔTns.

8. The system according to claim 5, characterized in that, The real-time measurement information module is divided into several parts, including: The submodule for classifying anomaly types is used to determine the range of anomaly nodes by analyzing the characteristic values ​​of the time-frequency nodes. By analyzing the evolution law of time-frequency parameters of a single node and the evolution law of time-frequency parameters of neighboring nodes in the cluster, the anomaly types of time-frequency nodes are classified into single-node time-frequency synchronization data packet loss fault, single-node synchronization network delay instability fault, single-node synchronization network networking instability fault, single-node time-frequency performance degradation exceeding threshold fault, single-node time-frequency performance sudden change fault, regional time-frequency synchronization data packet loss fault, regional synchronization network delay instability fault, regional synchronization network networking instability fault, and time-frequency equipment batch quality fault. Single-node time-frequency synchronization data packet loss fault refers to a situation where the node can collect monitoring data, but the node's last time was not updated, while the time-frequency characteristics of the lower-level node are consistent with this node, and the time-frequency characteristics of the upper-level node are normal. Single-node synchronous network latency instability fault refers to a node network latency standard deviation exceeding a threshold. A single-node synchronous network instability fault refers to a node being offline in the network. A single-node time-frequency performance degradation exceeding the threshold fault refers to a node frequency deviation aging rate exceeding the threshold, while the time-frequency characteristics of the upstream node are normal. A single-node time-frequency performance mutation fault refers to a comprehensive evaluation of the changes in the node's normalized frequency difference mean, node normalized frequency difference standard deviation, node time difference mean, and node time difference standard deviation. The evaluation result exceeds the threshold, while the time-frequency characteristics of the upper-level node are normal, and the time-frequency performance of the lower-level node is consistent with that node. Regional time-frequency synchronization data packet loss fault refers to data packet loss faults occurring at multiple nodes within the station area; Regional synchronization network latency instability fault refers to the network latency instability fault occurring at multiple nodes within the station area; The instability of the regional synchronization network indicates that multiple nodes within the station area are offline. Batch quality failure of time and frequency equipment refers to multiple devices in operation experiencing performance degradation exceeding the threshold or sudden changes in time and frequency performance.

Images