Power distribution terminal health state prediction method and device and electronic equipment

Abstract

The invention provides a power distribution terminal health state prediction method and device and electronic equipment, and belongs to the technical field of intelligent operation and maintenance of power distribution terminals. The method comprises the following steps: acquiring to-be-monitored multi-source monitoring data of a power distribution terminal; inputting the to-be-detected multi-source monitoring data into a power distribution terminal health state prediction model to obtain a power distribution terminal health state output by the power distribution terminal health state prediction model; the power distribution terminal health state prediction model is trained based on a source domain data set and a target domain data set by using a transfer learning algorithm; the source domain data set comprises sample multi-source monitoring data under a single working condition, and the target domain data set comprises sample multi-source monitoring data under multiple working conditions. The method is used for solving the problems that the evaluation dimension is single and the working condition adaptability is poor when an existing method is used for power distribution terminal health state early warning.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent operation and maintenance of distribution terminals, and specifically to a method for predicting the health status of a distribution terminal, a device for predicting the health status of a distribution terminal, an electronic device, a machine-readable storage medium, and a computer program product. Background Art

[0002] Sudden failures of key distribution terminals in distribution automation systems, such as distribution transformers and circuit breakers, can cause widespread power outages, significantly impacting people's lives and production. Therefore, health status early warnings for key distribution terminals, such as distribution transformers and circuit breakers, can help pinpoint the root cause of failures and mitigate the risk of sudden failures.

[0003] Current health status warnings for distribution terminals primarily rely on regular maintenance and single-parameter threshold assessments. Existing methods suffer from multiple limitations: First, they rely on a single evaluation dimension, relying solely on isolated parameters like temperature or current, while ignoring the synergistic impact of multiple data sources. Second, they lack adaptability to operating conditions. When terminals encounter sudden load changes or extreme weather conditions, traditional single-parameter threshold assessment methods are prone to significant deviations. Summary of the Invention

[0004] The purpose of the embodiments of the present invention is to provide a method, device and electronic device for predicting the health status of a distribution terminal, so as to solve the problems of single evaluation dimension and poor adaptability to working conditions when performing health status warning of a distribution terminal in existing methods.

[0005] To achieve the above objectives, an embodiment of the present invention provides a method for predicting the health status of a power distribution terminal, comprising: Obtain multi-source monitoring data to be tested at the power distribution terminal; Inputting the multi-source monitoring data to be tested into a power distribution terminal health status prediction model to obtain the power distribution terminal health status output by the power distribution terminal health status prediction model; Among them, the distribution terminal health status prediction model is trained based on a source domain dataset and a target domain dataset using a transfer learning algorithm; the source domain dataset includes sample multi-source monitoring data under a single working condition, and the target domain dataset includes sample multi-source monitoring data under multiple working conditions.

[0006] Optionally, the power distribution terminal health status prediction model is trained by the following steps: When the power distribution terminal switches its working condition, obtaining multi-source monitoring data of samples of the power distribution terminal under different working conditions; Constructing a joint training set based on the sample multi-source monitoring data; the joint training set includes the source domain dataset and the target domain dataset; Repeat all the following steps until the set end condition is met: calculate the sampling probability of each sample multi-source monitoring data in the joint training set; perform weighted random sampling on the joint training set based on the sampling probability to obtain a sub-sample data set; use the KD tree algorithm or the KNN algorithm to perform model training on the sub-sample data set to obtain a base classifier; calculate the error rate of the base classifier on the target domain data set; calculate the intermediate weight based on the error rate; adjust the sample weight of the source domain data set and the sample weight of the target domain data set based on the intermediate weight; constructing an optimized source domain dataset based on the sample weights of the source domain dataset after the set end condition is reached; The power distribution terminal health status prediction model is trained based on the optimized source domain data set.

[0007] Optionally, the using a KD tree algorithm to perform model training on the subsample dataset to obtain a base classifier includes: Dividing the subsample data set into q sub-areas; Constructing a KD tree for each sub-region in parallel; Based on the KD tree of each sub-region, parallel search is performed to find k nearest neighbor samples between each sub-region and the target domain dataset based on the Euclidean distance, and q*k candidate samples are obtained; Construct a target KD tree based on the q*k candidate samples; Searching for the k target nearest neighbor samples with the global minimum Euclidean distance to the target domain dataset based on the target KD tree; Model training is performed based on the k target nearest neighbor samples to obtain a base classifier.

[0008] Optionally, the using of the KNN algorithm to perform model training on the subsample dataset to obtain a base classifier includes: Dividing the subsample data set into q sub-areas; Calculating the Euclidean distance between the target domain dataset and the sample multi-source monitoring data in each of the sub-regions; Determine k initial nearest neighbor samples of each sub-region based on the Euclidean distance to obtain q*k candidate samples; Filter out the global minimum k target nearest neighbor samples from the q*k candidate samples; Model training is performed based on the k target nearest neighbor samples to obtain a base classifier.

[0009] Optionally, adjusting the sample weights of the source domain dataset and the sample weights of the target domain dataset based on the intermediate weights includes: When the classification of the sample multi-source monitoring data of the target domain dataset is correct, adjusting the sample weights of the sample multi-source monitoring data of the target domain dataset based on the intermediate weights so that the sample weights of the sample multi-source monitoring data of the target domain dataset are reduced; and when the classification of the sample multi-source monitoring data of the target domain dataset is incorrect, adjusting the sample weights of the sample multi-source monitoring data of the target domain dataset based on the intermediate weights so that the sample weights of the sample multi-source monitoring data of the target domain dataset remain unchanged; When the classification of the sample multi-source monitoring data of the source domain data set is correct, the sample weight of the sample multi-source monitoring data of the source domain data set is adjusted based on the intermediate weight so that the sample weight of the sample multi-source monitoring data of the source domain data set is increased; and when the classification of the sample multi-source monitoring data of the source domain data set is incorrect, the sample weight of the sample multi-source monitoring data of the source domain data set is adjusted based on the intermediate weight so that the sample weight of the sample multi-source monitoring data of the source domain data set is reduced.

[0010] Optionally, the multi-source monitoring data to be tested includes at least two of the electrical data, mechanical data and environmental data of the distribution terminal; the sample multi-source monitoring data includes at least two of the sample electrical data, sample mechanical data and sample environmental data of the distribution terminal.

[0011] On the other hand, an embodiment of the present invention further provides a device for predicting the health status of a power distribution terminal, comprising: An acquisition module is used to obtain the multi-source monitoring data to be tested of the power distribution terminal; A prediction module, configured to input the multi-source monitoring data to be tested into a power distribution terminal health status prediction model to obtain a power distribution terminal health status output by the power distribution terminal health status prediction model; Among them, the distribution terminal health status prediction model is trained based on a source domain dataset and a target domain dataset using a transfer learning algorithm; the source domain dataset includes sample multi-source monitoring data under a single working condition, and the target domain dataset includes sample multi-source monitoring data under multiple working conditions.

[0012] Optionally, the power distribution terminal health status prediction model is trained by the following steps: When the power distribution terminal switches its working condition, obtaining sample multi-source monitoring data of the power distribution terminal under different working conditions; Constructing a joint training set based on the sample multi-source monitoring data; the joint training set includes the source domain dataset and the target domain dataset; Repeat all the following steps until the set end condition is met: calculate the sampling probability of each sample multi-source monitoring data in the joint training set; perform weighted random sampling on the joint training set based on the sampling probability to obtain a sub-sample data set; use the KD tree algorithm or the KNN algorithm to perform model training on the sub-sample data set to obtain a base classifier; calculate the error rate of the base classifier on the target domain data set; calculate the intermediate weight based on the error rate; adjust the sample weight of the source domain data set and the sample weight of the target domain data set based on the intermediate weight; Constructing an optimized source domain dataset based on the sample weights of the source domain dataset after the set end condition is reached; The power distribution terminal health status prediction model is trained based on the optimized source domain data set.

[0013] Optionally, the using a KD tree algorithm to perform model training on the subsample dataset to obtain a base classifier includes: Dividing the subsample data set into q sub-areas; Constructing a KD tree for each sub-region in parallel; Based on the KD tree of each sub-region, parallel search is performed to find k nearest neighbor samples between each sub-region and the target domain dataset based on the Euclidean distance, and q*k candidate samples are obtained; Construct a target KD tree based on the q*k candidate samples; Searching for the k target nearest neighbor samples with the global minimum Euclidean distance to the target domain dataset based on the target KD tree; Model training is performed based on the k target nearest neighbor samples to obtain a base classifier.

[0014] Optionally, the using of the KNN algorithm to perform model training on the subsample dataset to obtain a base classifier includes: Dividing the subsample data set into q sub-areas; Calculating the Euclidean distance between the target domain dataset and the sample multi-source monitoring data in each of the sub-regions; Determine k initial nearest neighbor samples of each sub-region based on the Euclidean distance to obtain q*k candidate samples; Filter out the global minimum k target nearest neighbor samples from the q*k candidate samples; Model training is performed based on the k target nearest neighbor samples to obtain a base classifier.

[0015] Optionally, adjusting the sample weights of the source domain dataset and the sample weights of the target domain dataset based on the intermediate weights includes: When the classification of the sample multi-source monitoring data of the target domain dataset is correct, adjusting the sample weights of the sample multi-source monitoring data of the target domain dataset based on the intermediate weights so that the sample weights of the sample multi-source monitoring data of the target domain dataset are reduced; and when the classification of the sample multi-source monitoring data of the target domain dataset is incorrect, adjusting the sample weights of the sample multi-source monitoring data of the target domain dataset based on the intermediate weights so that the sample weights of the sample multi-source monitoring data of the target domain dataset remain unchanged; When the classification of the sample multi-source monitoring data of the source domain data set is correct, the sample weight of the sample multi-source monitoring data of the source domain data set is adjusted based on the intermediate weight so that the sample weight of the sample multi-source monitoring data of the source domain data set is increased; and when the classification of the sample multi-source monitoring data of the source domain data set is incorrect, the sample weight of the sample multi-source monitoring data of the source domain data set is adjusted based on the intermediate weight so that the sample weight of the sample multi-source monitoring data of the source domain data set is reduced.

[0016] Optionally, the multi-source monitoring data to be tested includes at least two of the electrical data, mechanical data and environmental data of the distribution terminal; the sample multi-source monitoring data includes at least two of the sample electrical data, sample mechanical data and sample environmental data of the distribution terminal.

[0017] On the other hand, the present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned method for predicting the health status of a power distribution terminal when executing the program.

[0018] On the other hand, the present invention also provides a machine-readable storage medium having a computer program stored thereon, which implements the above-mentioned method for predicting the health status of a power distribution terminal when executed by a processor.

[0019] On the other hand, the present invention also provides a computer program product, including a computer program, which implements the above-mentioned distribution terminal health status prediction method when executed by a processor.

[0020] Through the above technical solution, the health status prediction model of the distribution terminal in the embodiment of the present invention is trained based on sample multi-source monitoring data under multiple working conditions composed of source domain data sets and target domain data sets, avoiding the defect of a single evaluation dimension; and the health status prediction model of the distribution terminal in the embodiment of the present invention is based on the source domain data sets and the target domain data sets, and is trained using a transfer learning algorithm. It can screen source domain data that is closer to the distribution of the target domain data sets for training, suppress noise samples, and thus realize the health status prediction of the distribution terminal under multiple working conditions, solving the defect of poor adaptability to working conditions of the existing method.

[0021] Other features and advantages of the embodiments of the present invention will be described in detail in the subsequent detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The accompanying drawings are used to provide a further understanding of the embodiments of the present invention and constitute a part of the specification. Together with the following detailed description, they are used to explain the embodiments of the present invention, but do not constitute a limitation of the embodiments of the present invention. In the accompanying drawings: Figure 1 This is one of the flow charts of the method for predicting the health status of a power distribution terminal provided by the present invention; Figure 2 This is the second flow chart of the method for predicting the health status of a power distribution terminal provided by the present invention; Figure 3 It is a structural diagram of the device for predicting the health status of a power distribution terminal provided by the present invention; Figure 4 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0023] The following describes the specific implementation of the embodiment of the present invention in detail with reference to the accompanying drawings. It should be understood that the specific implementation described herein is only used to illustrate and explain the embodiment of the present invention and is not used to limit the embodiment of the present invention.

[0024] Method Example Please refer to Figure 1 The embodiment of the present invention provides a method for predicting the health status of a power distribution terminal, comprising: Step 100: Acquire multi-source monitoring data to be tested of the power distribution terminal.

[0025] The electronic device obtains the multi-source monitoring data to be tested of the distribution terminal. The distribution terminal can be various execution units in the distribution automation system. For example, in the embodiment of the present invention, the distribution terminal can be a feeder terminal (FTU, Feeder Terminal Unit), a station terminal (DTU, Distribution Terminal Unit), a distribution transformer terminal (TTU, Transformer Terminal Unit), or a low-voltage monitoring terminal (LTU, Low-voltage Terminal Unit), etc. The multi-source monitoring data to be tested may include at least two of the electrical data, mechanical data and environmental data of the distribution terminal. That is, the multi-source monitoring data to be tested may include any two or three of the electrical data, mechanical data and environmental data of the distribution terminal.

[0026] In order to expand the evaluation dimension, the synergistic influence of multi-source data is considered as much as possible, and in order to improve the dynamic feature fusion capability, the heterogeneous information generated during the operation of the distribution terminal is effectively integrated. In one embodiment, the multi-source monitoring data to be tested may include the electrical data, mechanical data and environmental data of the distribution terminal. The electrical data of the distribution terminal may include data such as the voltage, current, power and harmonics of the distribution terminal. The mechanical data of the distribution terminal may include data such as the vibration data and structural protection level of the distribution terminal. The environmental data of the distribution terminal may include data such as the temperature and humidity of the environment in which the distribution terminal is located.

[0027] Step 200: Input the multi-source monitoring data to be tested into a power distribution terminal health status prediction model to obtain the power distribution terminal health status output by the power distribution terminal health status prediction model.

[0028] The electronic device inputs the multi-source monitoring data to be tested into the distribution terminal health status prediction model to obtain the distribution terminal health status output by the distribution terminal health status prediction model. The distribution terminal health status prediction model is trained based on a source domain data set and a target domain data set using a transfer learning algorithm; the source domain data set includes sample multi-source monitoring data under a single working condition, and the target domain data set includes sample multi-source monitoring data under multiple working conditions. In the embodiment of the present invention, the source domain data set can be understood as sample multi-source monitoring data under a single working condition in the old data of the distribution terminal. The target domain data set can be understood as sample multi-source monitoring data under multiple working conditions in the new data of the distribution terminal.

[0029] In one embodiment, the multiple operating conditions of the distribution terminal may include heavy load conditions, light load conditions, and impact load conditions of the distribution terminal. The light load condition may be when the actual load of the distribution terminal is far below the rated capacity, for example, the load rate is generally ≤30%. The heavy load condition may be when the actual load rate of the distribution terminal is between 70% and 90%, close to the rated capacity. The impact load condition may be a non-periodic or periodic sudden load, with a peak value of up to dozens of times the average load. The sample multi-source monitoring data includes at least two of the sample electrical data, sample mechanical data, and sample environmental data of the distribution terminal. That is, the sample multi-source monitoring data may include any two or three of the sample electrical data, sample mechanical data, and sample environmental data of the distribution terminal. In one embodiment, the sample multi-source monitoring data may include the sample electrical data, sample mechanical data, and sample environmental data of the distribution terminal. The sample multi-source monitoring data may be obtained based on the historical multi-source monitoring data of the distribution terminal. The health status prediction model of the distribution terminal may use various neural network models, such as convolutional neural networks, long short-term memory networks, etc.

[0030] Because when the operating conditions of the distribution terminal change, the existing training data set under a single operating condition is difficult to include all potential operating conditions. And when the distribution difference between the target domain and the source domain is large, the algorithm is susceptible to noise interference and the effect is poor. Therefore, in order to make the data distribution of the source domain data set closer to the distribution of the target domain data set, suppress noise samples, and realize the health status prediction of the distribution terminal under multiple working conditions. The distribution terminal health status prediction model (or distribution equipment health status warning model) of the embodiment of the present invention can be trained based on the source domain data set and the target domain data set using a transfer learning algorithm. The goal of the transfer learning algorithm is to effectively extract information from the knowledge learned in one environment to help learning in a new environment. For example, the embodiment of the present invention can use the TrAdaBoost algorithm to fine-tune the model. TrAdaBoost is an instance-based transfer learning algorithm with strong knowledge transfer capabilities. When the operating conditions of the distribution terminal change, the existing training data set is difficult to include all potential operating conditions. The embodiment of the present invention is based on the model fine-tuning method of improving TrAdaBoost, which can screen source domain data that is closer to the distribution of the target domain data set for training and suppress noise samples, thereby realizing the health status prediction of the distribution terminal under multiple operating conditions and solving the defect of poor adaptability of existing methods to working conditions.

[0031] Therefore, the health status prediction model of the distribution terminal in the embodiment of the present invention is trained based on sample multi-source monitoring data under multiple working conditions composed of source domain data sets and target domain data sets, avoiding the defect of a single evaluation dimension; and the health status prediction model of the distribution terminal in the embodiment of the present invention is trained based on the source domain data sets and target domain data sets using a transfer learning algorithm, which can screen source domain data that is closer to the distribution of the target domain data set for training, suppress noise samples, and thus realize the health status prediction of the distribution terminal under multiple working conditions, solving the defect of poor adaptability to working conditions of the existing method.

[0032] For other aspects of the present invention, please refer to Figure 2 The power distribution terminal health status prediction model is trained by the following steps: Step 11: When the power distribution terminal switches its working condition, obtain sample multi-source monitoring data of the power distribution terminal under different working conditions.

[0033] When the power distribution terminal switches its working condition, the electronic device obtains sample multi-source monitoring data of the power distribution terminal under different working conditions.

[0034] Step 12: construct a joint training set based on the sample multi-source monitoring data; the joint training set includes the source domain dataset and the target domain dataset.

[0035] Electronic equipment collects multi-source monitoring data of distribution terminals under different working conditions and constructs a joint training set ,T a is the source domain dataset, T b is the target domain dataset. And the electronic device can initialize the sample weights in the source domain dataset and the target domain dataset: the sample weight of the source domain dataset is ; The sample weights in the target domain dataset are Where n and m are the number of samples, i represents the multi-source monitoring data of the i-th sample in the source domain dataset, and j represents the multi-source monitoring data of the j-th sample in the target domain dataset.

[0036] Step 13, repeat all the following steps until the set end condition is reached: calculate the sampling probability of each sample multi-source monitoring data in the joint training set; perform weighted random sampling on the joint training set based on the sampling probability to obtain a sub-sample data set; use the KD tree algorithm or the KNN algorithm to perform model training on the sub-sample data set to obtain a base classifier; calculate the error rate of the base classifier on the target domain data set; calculate the intermediate weight based on the error rate; adjust the sample weight of the source domain data set and the sample weight of the target domain data set based on the intermediate weight.

[0037] The electronic device can initialize the weight vector w 1 =(w 1 1,…,w 1 n+m ).in, ; Where n is the number of samples in the source domain dataset, and m is the number of samples in the target domain dataset. The electronic device calculates the sampling probability of each sample multi-source monitoring data in the joint training set as ; where t is the number of iterations. The sampling probability of electronic equipment p t The joint training set is subjected to weighted random sampling to obtain a subsample data set M. The electronic device can divide the subsample data set M into q sub-areas, and each sub-area independently calls the KD tree algorithm or the KNN algorithm to perform model training on the subsample data set to obtain a base classifier h t . The embodiment of the present invention proposes an improved model fine-tuning method of TrAdaBoost, designs a parallel multi-channel method, divides the sub-sample data set M into multiple sub-regions (parallel channels), and uses the KD tree algorithm or the KNN algorithm in parallel for each sub-region to accelerate model training. In addition, the embodiment of the present invention proposes to adopt a weighted random sampling method to dynamically adjust the sampling probability according to the sample weight, screen the source domain data that is closer to the target domain distribution, and suppress noise samples. The electronic device then calculates the error rate of the base classifier on the target domain data set. .in ≥0.5, terminate the iteration. Represents an indicator function that outputs 1 if the condition in the brackets is met, otherwise 0. For example, The value is 1 if the classification is incorrect. x t j represents the jth sample multi-source monitoring data in the target domain dataset, and t represents the number of iterations. t ( x t j ) represents the base classifier h t right x t j The health status prediction result, y t j express x t j The actual health status of the corresponding distribution terminal. represents the sample weight of the jth sample in the target domain dataset at the tth iteration. The electronic device then calculates the intermediate weight based on the error rate. In one embodiment, the electronic device calculates the intermediate weight based on the error rate. Update intermediate weights β t and γ t Among them, the intermediate weight β t and γ t The calculation formula is as follows: ; The electronic device then adjusts the sample weights of the source domain dataset and the sample weights of the target domain dataset based on the intermediate weights. Repeat the above steps until the value is reached. The termination condition of iteration is ≥0.5.

[0038] Step 14: construct an optimized source domain dataset based on the sample weights of the source domain dataset after the set end condition is reached.

[0039] Step 15: Training the power distribution terminal health status prediction model based on the optimized source domain dataset.

[0040] The electronic device normalizes the sample weights of the source domain data set after the set end condition is reached to obtain an optimized source domain data set. The electronic device then trains the distribution terminal health status prediction model based on the optimized source domain data set. Specifically, the electronic device can be trained in a supervised manner based on the optimized source domain data set, calculate the loss function by calculating the deviation between the model prediction result and the true label, and adjust the model parameters of the distribution terminal health status prediction model based on the loss function to obtain a trained distribution terminal health status prediction model. The embodiment of the present invention updates the distribution terminal health status prediction model in real time when the distribution terminal switches the working condition, thereby improving the generalization ability of the model.

[0041] When the data distribution of the target domain dataset differs significantly from that of the source domain dataset, the algorithm is susceptible to noise interference and performs poorly. The model fine-tuning method based on the improved TrAdaBoost designed in this embodiment of the present invention first uses the KNN algorithm or KD tree as the learner and designs a parallel multi-channel approach to divide the subsample dataset into multiple sub-regions (parallel channels) to accelerate model training. It also proposes a weighted random sampling approach that dynamically adjusts the sampling probability based on sample weights, selects source domain data that more closely matches the target domain distribution, and suppresses noisy samples.

[0042] In other aspects of the embodiments of the present invention, the use of the KD tree algorithm to perform model training on the sub-sample dataset to obtain a base classifier includes: dividing the sub-sample dataset into q sub-regions; constructing a KD tree for each of the sub-regions in parallel; based on the KD tree of each of the sub-regions, searching in parallel for k nearest neighbor samples based on the Euclidean distance between each of the sub-regions and the target domain dataset to obtain q*k candidate samples; constructing a target KD tree based on the q*k candidate samples; searching for k target nearest neighbor samples with the global minimum Euclidean distance to the target domain dataset based on the target KD tree; and performing model training based on the k target nearest neighbor samples to obtain a base classifier.

[0043] The electronic device divides the subsample data set M into q sub-regions, independently constructs a KD tree for the data of each sub-region, searches all KD trees in parallel during query, and takes the k nearest neighbor samples based on the Euclidean distance with the target domain data set after merging the results. Specifically, the electronic device constructs the KD tree for each sub-region in parallel during the construction phase. In the query phase, based on the KD tree of each sub-region, the k nearest neighbor samples based on the Euclidean distance between each sub-region and the target domain data set are searched in parallel to obtain q*k candidate samples; finally, the electronic device merges the q*k candidate samples into a large set, constructs a target KD tree, and searches for the k target nearest neighbor samples with the global minimum Euclidean distance with the target domain data set based on the target KD tree. Finally, the model is trained based on the k target nearest neighbor samples to obtain a base classifier.

[0044] Therefore, the embodiment of the present invention solves the problem of low computing efficiency of the power distribution terminal when the data distribution difference between the source domain dataset and the target domain dataset is large under multiple working conditions through weighted random sampling and parallel computing using the KD tree algorithm.

[0045] In other aspects of the embodiments of the present invention, the use of the KNN algorithm to perform model training on the sub-sample data set to obtain a base classifier includes: dividing the sub-sample data set into q sub-areas; calculating the Euclidean distance between the target domain data set and the sample multi-source monitoring data in each of the sub-areas; determining k initial nearest neighbor samples for each of the sub-areas based on the Euclidean distance to obtain q*k candidate samples; screening out the globally minimum k target nearest neighbor samples from the q*k candidate samples; and performing model training based on the k target nearest neighbor samples to obtain a base classifier.

[0046] The electronic device first uses the KNN algorithm as a learner and designs a parallel multi-channel approach to divide the subsample dataset into multiple sub-regions (parallel channels). The Euclidean distance between each sub-region and the target domain dataset is calculated and then the results are merged to accelerate model training. Specifically, the subsample dataset is divided into q sub-regions; the Euclidean distance between the target domain dataset and the multi-source monitoring data of samples in each sub-region is calculated; based on the Euclidean distance, k initial nearest neighbor samples are determined for each sub-region, resulting in q*k candidate samples; the globally minimum k target nearest neighbor samples are screened from these q*k candidate samples; and the model is trained based on these k target nearest neighbor samples to obtain a base classifier.

[0047] Therefore, the embodiment of the present invention solves the problem of low computing efficiency of the power distribution terminal when the data distribution difference between the source domain dataset and the target domain dataset is large under multiple working conditions through weighted random sampling and parallel computing using the KNN algorithm.

[0048] In other aspects of the embodiments of the present invention, adjusting the sample weights of the source domain dataset and the sample weights of the target domain dataset based on the intermediate weights includes: When the classification of the sample multi-source monitoring data of the target domain dataset is correct, the sample weight of the sample multi-source monitoring data of the target domain dataset is adjusted based on the intermediate weight so that the sample weight of the sample multi-source monitoring data of the target domain dataset is reduced; and when the classification of the sample multi-source monitoring data of the target domain dataset is incorrect, the sample weight of the sample multi-source monitoring data of the target domain dataset is adjusted based on the intermediate weight so that the sample weight of the sample multi-source monitoring data of the target domain dataset remains unchanged.

[0049] In the embodiment of the present invention, the intermediate weight β t The calculation formula is as follows: ;in, is the error rate of the base classifier on the target domain dataset, β t is an intermediate weight. When the classification of the sample multi-source monitoring data of the target domain data set is correct, the sample weight of the sample multi-source monitoring data of the target domain data set is adjusted based on the intermediate weight, so that the sample weight of the sample multi-source monitoring data of the target domain data set is reduced; and when the classification of the sample multi-source monitoring data of the target domain data set is incorrect, the sample weight of the sample multi-source monitoring data of the target domain data set is adjusted based on the intermediate weight, so that the sample weight of the sample multi-source monitoring data of the target domain data set remains unchanged. Specifically, the sample weight update formula of the sample multi-source monitoring data of the target domain data set is as follows: ; in, w b j ( t +1) is the t +1 times the sample weight of the j-th sample multi-source monitoring data, w b j ( t ) is the t The sample weight of the j-th sample multi-source monitoring data; β tis the intermediate weight; t represents the number of iterations. The embodiment of the present invention achieves that for the target domain dataset, the weight of correctly classified samples is reduced, while the weight of erroneous samples remains unchanged. Therefore, the embodiment of the present invention prevents the erroneous samples of the target domain dataset from being ignored due to the weight reduction by making the weight remain unchanged, forcing the subsequent base classifier to continuously optimize these difficult samples, and improving the generalization ability by retaining the difficult samples. The embodiment of the present invention reduces the redundant calculation of the model on simple samples by reducing the weight of the correct samples of the target domain dataset, that is, reduces redundant learning, thereby focusing on difficult samples. Therefore, the embodiment of the present invention accelerates the learning of difficult samples in the target domain and avoids underfitting.

[0050] When the classification of the sample multi-source monitoring data of the source domain data set is correct, the sample weight of the sample multi-source monitoring data of the source domain data set is adjusted based on the intermediate weight so that the sample weight of the sample multi-source monitoring data of the source domain data set is increased; and when the classification of the sample multi-source monitoring data of the source domain data set is incorrect, the sample weight of the sample multi-source monitoring data of the source domain data set is adjusted based on the intermediate weight so that the sample weight of the sample multi-source monitoring data of the source domain data set is reduced.

[0051] In the embodiment of the present invention, the intermediate weight β t The calculation formula is as follows: ;in, is the error rate of the base classifier on the target domain dataset, β t is an intermediate weight. When the classification of the sample multi-source monitoring data of the source domain data set is correct, the sample weight of the sample multi-source monitoring data of the source domain data set is adjusted based on the intermediate weight, so that the sample weight of the sample multi-source monitoring data of the source domain data set is increased; and when the classification of the sample multi-source monitoring data of the source domain data set is incorrect, the sample weight of the sample multi-source monitoring data of the source domain data set is adjusted based on the intermediate weight, so that the sample weight of the sample multi-source monitoring data of the source domain data set is reduced. Specifically, the sample weight update formula of the sample multi-source monitoring data of the source domain data set is as follows: ; in, For the t +1 times the sample weight of the i-th sample multi-source monitoring data, For the t The sample weight of the i-th sample multi-source monitoring data; β t and γ tis the intermediate weight; t represents the number of iterations. Thus, the embodiment of the present invention achieves that for the source domain dataset, the weight of correctly classified samples is increased, and the weight of incorrect samples is reduced. Thus, the embodiment of the present invention reduces the weight of incorrectly classified samples in the source domain dataset, gradually eliminates source domain samples with large distribution differences from the target domain dataset, and filters out interfering samples with distribution drift, thereby suppressing noise with inconsistent distribution. The embodiment of the present invention increases the weight of correctly classified samples in the source domain dataset, retains transferable common knowledge, and improves the utilization rate of the source domain dataset. That is, effective knowledge is retained and knowledge related to the target domain dataset is strengthened. Thus, the embodiment of the present invention achieves accurate screening of knowledge in the source domain dataset.

[0052] In other aspects of the embodiment of the present invention, after step 200, the process further includes sending an alarm signal to the operation and maintenance system based on the abnormal health status of the power distribution terminal, thereby achieving health status early warning of the power distribution terminal under multiple working conditions and facilitating timely locating the root cause of the fault.

[0053] In summary, the embodiment of the present invention is based on the model fine-tuning method of the improved transfer learning algorithm, and innovatively introduces a parallel multi-channel weighted sampling mechanism and a dynamic distribution adaptation strategy to realize the health status prediction of the distribution terminal under multiple working conditions. The embodiment of the present invention can use the historical multi-source monitoring data of the distribution terminal based on cross-domain knowledge transfer to quickly predict the health status of the distribution terminal, thereby solving the problems of computational efficiency and model generalization.

[0054] Device embodiment Please refer to Figure 3 On the other hand, an embodiment of the present invention further provides a device for predicting the health status of a power distribution terminal, comprising: An acquisition module 301 is used to acquire the multi-source monitoring data to be tested of the power distribution terminal; The prediction module 302 is configured to input the multi-source monitoring data to be tested into a power distribution terminal health status prediction model to obtain the power distribution terminal health status output by the power distribution terminal health status prediction model; Among them, the distribution terminal health status prediction model is trained based on a source domain dataset and a target domain dataset using a transfer learning algorithm; the source domain dataset includes sample multi-source monitoring data under a single working condition, and the target domain dataset includes sample multi-source monitoring data under multiple working conditions.

[0055] Optionally, the power distribution terminal health status prediction model is trained by the following steps: When the power distribution terminal switches its working condition, obtaining multi-source monitoring data of samples of the power distribution terminal under different working conditions; Constructing a joint training set based on the sample multi-source monitoring data; the joint training set includes the source domain dataset and the target domain dataset; Repeat all the following steps until the set end condition is met: calculate the sampling probability of each sample multi-source monitoring data in the joint training set; perform weighted random sampling on the joint training set based on the sampling probability to obtain a sub-sample data set; use the KD tree algorithm or the KNN algorithm to perform model training on the sub-sample data set to obtain a base classifier; calculate the error rate of the base classifier on the target domain data set; calculate the intermediate weight based on the error rate; adjust the sample weight of the source domain data set and the sample weight of the target domain data set based on the intermediate weight; constructing an optimized source domain dataset based on the sample weights of the source domain dataset after the set end condition is reached; The power distribution terminal health status prediction model is trained based on the optimized source domain data set.

[0056] Optionally, the using a KD tree algorithm to perform model training on the subsample dataset to obtain a base classifier includes: Dividing the subsample data set into q sub-areas; Constructing a KD tree for each sub-region in parallel; Based on the KD tree of each sub-region, parallel search is performed to find k nearest neighbor samples between each sub-region and the target domain dataset based on the Euclidean distance, and q*k candidate samples are obtained; Construct a target KD tree based on the q*k candidate samples; Searching for the k target nearest neighbor samples with the global minimum Euclidean distance to the target domain dataset based on the target KD tree; Model training is performed based on the k target nearest neighbor samples to obtain a base classifier.

[0057] Optionally, the using of the KNN algorithm to perform model training on the subsample dataset to obtain a base classifier includes: Dividing the subsample data set into q sub-areas; Calculating the Euclidean distance between the target domain dataset and the sample multi-source monitoring data in each of the sub-regions; Determine k initial nearest neighbor samples of each sub-region based on the Euclidean distance to obtain q*k candidate samples; Filter out the global minimum k target nearest neighbor samples from the q*k candidate samples; Model training is performed based on the k target nearest neighbor samples to obtain a base classifier.

[0058] Optionally, adjusting the sample weights of the source domain dataset and the sample weights of the target domain dataset based on the intermediate weights includes: When the classification of the sample multi-source monitoring data of the target domain dataset is correct, adjusting the sample weights of the sample multi-source monitoring data of the target domain dataset based on the intermediate weights so that the sample weights of the sample multi-source monitoring data of the target domain dataset are reduced; and when the classification of the sample multi-source monitoring data of the target domain dataset is incorrect, adjusting the sample weights of the sample multi-source monitoring data of the target domain dataset based on the intermediate weights so that the sample weights of the sample multi-source monitoring data of the target domain dataset remain unchanged; When the classification of the sample multi-source monitoring data of the source domain data set is correct, the sample weight of the sample multi-source monitoring data of the source domain data set is adjusted based on the intermediate weight so that the sample weight of the sample multi-source monitoring data of the source domain data set is increased; and when the classification of the sample multi-source monitoring data of the source domain data set is incorrect, the sample weight of the sample multi-source monitoring data of the source domain data set is adjusted based on the intermediate weight so that the sample weight of the sample multi-source monitoring data of the source domain data set is reduced.

[0059] Optionally, the multi-source monitoring data to be tested includes at least two of the electrical data, mechanical data and environmental data of the distribution terminal; the sample multi-source monitoring data includes at least two of the sample electrical data, sample mechanical data and sample environmental data of the distribution terminal.

[0060] The power distribution terminal health status prediction device includes a processor and a memory. The acquisition module 301, prediction module 302, etc. are stored in the memory as program units, and the processor executes the program units stored in the memory to implement corresponding functions.

[0061] The processor includes a kernel, which calls the corresponding program unit from the memory. There can be one or more kernels.

[0062] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0063] Figure 4 An example of a physical structure diagram of an electronic device is shown below. Figure 4As shown, the electronic device may include: a processor (processor) 410, a communication interface (Communications Interface) 420, a memory (memory) 430 and a communication bus 440, wherein the processor 410, the communication interface 420, and the memory 430 communicate with each other via the communication bus 440. The processor 410 may call the logic instructions in the memory 430 to execute a method for predicting the health status of a distribution terminal, the method comprising: obtaining multi-source monitoring data to be tested of the distribution terminal; inputting the multi-source monitoring data to be tested into a distribution terminal health status prediction model to obtain the health status of the distribution terminal output by the distribution terminal health status prediction model; wherein the distribution terminal health status prediction model is trained based on a source domain data set and a target domain data set using a transfer learning algorithm; the source domain data set includes sample multi-source monitoring data under a single working condition, and the target domain data set includes sample multi-source monitoring data under multiple working conditions.

[0064] Furthermore, the logic instructions in the aforementioned memory 430 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product, stored in a storage medium, includes instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0065] On the other hand, the present invention also provides a computer program product, which includes a computer program, which can be stored on a machine-readable storage medium. When the computer program is executed by a processor, the computer can execute a distribution terminal health status prediction method, which includes: obtaining multi-source monitoring data to be tested of the distribution terminal; inputting the multi-source monitoring data to be tested into a distribution terminal health status prediction model to obtain the distribution terminal health status output by the distribution terminal health status prediction model; wherein the distribution terminal health status prediction model is trained based on a source domain data set and a target domain data set using a transfer learning algorithm; the source domain data set includes sample multi-source monitoring data under a single working condition, and the target domain data set includes sample multi-source monitoring data under multiple working conditions.

[0066] On the other hand, the present invention also provides a machine-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute a distribution terminal health status prediction method, the method comprising: obtaining multi-source monitoring data to be tested of the distribution terminal; inputting the multi-source monitoring data to be tested into a distribution terminal health status prediction model to obtain the distribution terminal health status output by the distribution terminal health status prediction model; wherein the distribution terminal health status prediction model is trained based on a source domain data set and a target domain data set using a transfer learning algorithm; the source domain data set includes sample multi-source monitoring data under a single working condition, and the target domain data set includes sample multi-source monitoring data under multiple working conditions.

[0067] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0068] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.

[0069] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. 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 various embodiments of the present invention.

Claims

1. A method for predicting the health status of a power distribution terminal, characterized in that: include: Obtain multi-source monitoring data to be tested at the power distribution terminal; Inputting the multi-source monitoring data to be tested into a power distribution terminal health status prediction model to obtain the power distribution terminal health status output by the power distribution terminal health status prediction model; Among them, the distribution terminal health status prediction model is trained based on a source domain dataset and a target domain dataset using a transfer learning algorithm; the source domain dataset includes sample multi-source monitoring data under a single working condition, and the target domain dataset includes sample multi-source monitoring data under multiple working conditions.

2. The method for predicting the health status of a power distribution terminal according to claim 1, characterized in that: The power distribution terminal health status prediction model is trained by the following steps: When the power distribution terminal switches its working condition, obtaining multi-source monitoring data of samples of the power distribution terminal under different working conditions; Constructing a joint training set based on the sample multi-source monitoring data; the joint training set includes the source domain dataset and the target domain dataset; Repeat all the following steps until the set end condition is reached: Calculate the sampling probability of each sample multi-source monitoring data in the joint training set; Performing weighted random sampling on the joint training set based on the sampling probability to obtain a subsample data set; Performing model training on the subsample data set using a KD tree algorithm or a KNN algorithm to obtain a base classifier; Calculating the error rate of the base classifier on the target domain data set; calculating an intermediate weight based on the error rate; Adjusting the sample weights of the source domain dataset and the sample weights of the target domain dataset based on the intermediate weights; constructing an optimized source domain dataset based on the sample weights of the source domain dataset after the set end condition is reached; The power distribution terminal health status prediction model is trained based on the optimized source domain data set.

3. The method for predicting the health status of a power distribution terminal according to claim 2, characterized in that: The method of using the KD tree algorithm to perform model training on the subsample data set to obtain a base classifier includes: Dividing the subsample data set into q sub-areas; Constructing a KD tree for each sub-region in parallel; Based on the KD tree of each sub-region, parallel search is performed to find k nearest neighbor samples between each sub-region and the target domain dataset based on the Euclidean distance, and q*k candidate samples are obtained; Construct a target KD tree based on the q*k candidate samples; Searching for the k target nearest neighbor samples with the global minimum Euclidean distance to the target domain dataset based on the target KD tree; Model training is performed based on the k target nearest neighbor samples to obtain a base classifier.

4. The method for predicting the health status of a power distribution terminal according to claim 2, wherein: The method of using the KNN algorithm to perform model training on the subsample data set to obtain a base classifier includes: Dividing the subsample data set into q sub-areas; Calculating the Euclidean distance between the target domain dataset and the sample multi-source monitoring data in each of the sub-regions; Determine k initial nearest neighbor samples of each sub-region based on the Euclidean distance to obtain q*k candidate samples; Filter out the global minimum k target nearest neighbor samples from the q*k candidate samples; Model training is performed based on the k target nearest neighbor samples to obtain a base classifier.

5. The method for predicting the health status of a power distribution terminal according to claim 2, characterized in that: The adjusting the sample weights of the source domain dataset and the sample weights of the target domain dataset based on the intermediate weights includes: When the classification of the sample multi-source monitoring data of the target domain dataset is correct, adjusting the sample weights of the sample multi-source monitoring data of the target domain dataset based on the intermediate weights so that the sample weights of the sample multi-source monitoring data of the target domain dataset are reduced; and when the classification of the sample multi-source monitoring data of the target domain dataset is incorrect, adjusting the sample weights of the sample multi-source monitoring data of the target domain dataset based on the intermediate weights so that the sample weights of the sample multi-source monitoring data of the target domain dataset remain unchanged; When the classification of the sample multi-source monitoring data of the source domain data set is correct, the sample weight of the sample multi-source monitoring data of the source domain data set is adjusted based on the intermediate weight so that the sample weight of the sample multi-source monitoring data of the source domain data set is increased; and when the classification of the sample multi-source monitoring data of the source domain data set is incorrect, the sample weight of the sample multi-source monitoring data of the source domain data set is adjusted based on the intermediate weight so that the sample weight of the sample multi-source monitoring data of the source domain data set is reduced.

6. The method for predicting the health status of a power distribution terminal according to any one of claims 1 to 5, characterized in that: The multi-source monitoring data to be tested includes at least two of the electrical data, mechanical data and environmental data of the distribution terminal; the sample multi-source monitoring data includes at least two of the sample electrical data, sample mechanical data and sample environmental data of the distribution terminal.

7. A device for predicting the health status of a power distribution terminal, characterized in that: include: An acquisition module is used to obtain the multi-source monitoring data to be tested of the power distribution terminal; A prediction module, configured to input the multi-source monitoring data to be tested into a power distribution terminal health status prediction model to obtain a power distribution terminal health status output by the power distribution terminal health status prediction model; Among them, the distribution terminal health status prediction model is trained based on a source domain dataset and a target domain dataset using a transfer learning algorithm; the source domain dataset includes sample multi-source monitoring data under a single working condition, and the target domain dataset includes sample multi-source monitoring data under multiple working conditions.

8. The device for predicting the health status of a power distribution terminal according to claim 7, characterized in that: The power distribution terminal health status prediction model is trained by the following steps: When the power distribution terminal switches its working condition, obtaining multi-source monitoring data of samples of the power distribution terminal under different working conditions; Constructing a joint training set based on the sample multi-source monitoring data; the joint training set includes the source domain dataset and the target domain dataset; Repeat all the following steps until the set end condition is reached: Calculate the sampling probability of each sample multi-source monitoring data in the joint training set; Performing weighted random sampling on the joint training set based on the sampling probability to obtain a subsample data set; Performing model training on the subsample data set using a KD tree algorithm or a KNN algorithm to obtain a base classifier; Calculating the error rate of the base classifier on the target domain data set; calculating an intermediate weight based on the error rate; Adjusting the sample weights of the source domain dataset and the sample weights of the target domain dataset based on the intermediate weights; constructing an optimized source domain dataset based on the sample weights of the source domain dataset after the set end condition is reached; The power distribution terminal health status prediction model is trained based on the optimized source domain data set.

9. The device for predicting the health status of a power distribution terminal according to claim 8, characterized in that: The method of using the KD tree algorithm to perform model training on the subsample data set to obtain a base classifier includes: Dividing the subsample data set into q sub-areas; Constructing a KD tree for each sub-region in parallel; Based on the KD tree of each sub-region, parallel search is performed to find k nearest neighbor samples between each sub-region and the target domain dataset based on the Euclidean distance, and q*k candidate samples are obtained; Construct a target KD tree based on the q*k candidate samples; Searching for the k target nearest neighbor samples with the global minimum Euclidean distance to the target domain dataset based on the target KD tree; Model training is performed based on the k target nearest neighbor samples to obtain a base classifier.

10. The device for predicting the health status of a power distribution terminal according to claim 8, characterized in that: The method of using the KNN algorithm to perform model training on the subsample data set to obtain a base classifier includes: Dividing the subsample data set into q sub-areas; Calculating the Euclidean distance between the target domain dataset and the sample multi-source monitoring data in each of the sub-regions; Determine k initial nearest neighbor samples of each sub-region based on the Euclidean distance to obtain q*k candidate samples; Filter out the global minimum k target nearest neighbor samples from the q*k candidate samples; Model training is performed based on the k target nearest neighbor samples to obtain a base classifier.

11. The device for predicting the health status of a power distribution terminal according to claim 8, characterized in that: The adjusting the sample weights of the source domain dataset and the sample weights of the target domain dataset based on the intermediate weights includes: When the classification of the sample multi-source monitoring data of the target domain dataset is correct, adjusting the sample weights of the sample multi-source monitoring data of the target domain dataset based on the intermediate weights so that the sample weights of the sample multi-source monitoring data of the target domain dataset are reduced; and when the classification of the sample multi-source monitoring data of the target domain dataset is incorrect, adjusting the sample weights of the sample multi-source monitoring data of the target domain dataset based on the intermediate weights so that the sample weights of the sample multi-source monitoring data of the target domain dataset remain unchanged; When the classification of the sample multi-source monitoring data of the source domain data set is correct, the sample weight of the sample multi-source monitoring data of the source domain data set is adjusted based on the intermediate weight so that the sample weight of the sample multi-source monitoring data of the source domain data set is increased; and when the classification of the sample multi-source monitoring data of the source domain data set is incorrect, the sample weight of the sample multi-source monitoring data of the source domain data set is adjusted based on the intermediate weight so that the sample weight of the sample multi-source monitoring data of the source domain data set is reduced.

12. The device for predicting the health status of a power distribution terminal according to any one of claims 7 to 11, characterized in that: The multi-source monitoring data to be tested includes at least two of the electrical data, mechanical data and environmental data of the distribution terminal; the sample multi-source monitoring data includes at least two of the sample electrical data, sample mechanical data and sample environmental data of the distribution terminal.

13. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the method for predicting the health status of a power distribution terminal according to any one of claims 1 to 6 is implemented.

14. A machine-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for predicting the health status of a power distribution terminal according to any one of claims 1 to 6 is implemented.

15. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the method for predicting the health status of a power distribution terminal according to any one of claims 1 to 6 is implemented.

Images