Power grid operation time sequence behavior identification method and system based on continuous learning

Abstract

The invention relates to the technical field of electric power operation behavior recognition, in particular to a power grid operation time sequence behavior recognition method and system based on continuous learning. The method comprises the following steps: collecting 1080P high-definition video data of a power grid power failure operation scene, covering different illumination, time, position and weather conditions, and carrying out key frame and time period selection and category labeling by professionals; a target detection model is adopted to generate a detection frame label, and a double-flow feature extraction network is utilized to obtain spatial-temporal features; a multi-strategy sample selection module is constructed, challenging samples are selected through error change rate and similarity calculation among samples, and a sample memory bank is established; based on a continuous learning principle, combining the samples of the two task periods to carry out iterative training; and finally, identification of specific time sequence behaviors such as climbing and grounding wire hanging is realized. According to the method, through continuous learning and multi-strategy sample selection, the recognition accuracy is remarkably improved, precise monitoring of a complex power grid operation scene is realized, and an effective guarantee is provided for safe operation of a power system.

Description

Technical Field

[0001] The present invention relates to the technical field of electric power operation behavior recognition, and in particular to a method and system for recognizing power grid operation timing behavior based on continuous learning. Background Art

[0002] With the rapid development of modern power grid systems, power grid operation scenarios are becoming more complex and diverse, and the behavior of operators directly affects the safety, stability and operating efficiency of the power system. In order to ensure the efficient operation and operational safety of the power grid, how to identify, monitor and respond to the behavior of operators in real time has become a critical technical problem in the current power grid field. Traditional behavior recognition methods usually rely on qualitative analysis and manual monitoring, which have significant limitations, especially when facing complex and changeable power grid operation scenarios, with low efficiency and delayed response. In addition, in the event of sudden failures or abnormal conditions, traditional methods cannot quickly identify potential risks, which increases the safety risks and operating pressure of the system.

[0003] In recent years, with the vigorous development of technologies such as computer vision, deep learning, and sensor fusion, intelligent behavior recognition technology based on video data and multi-source sensor information has gradually become the mainstream solution for monitoring power grid operation behavior. Compared with traditional methods, machine learning models can more quickly and accurately identify the behavior patterns of operators by automatically extracting features, especially in the detection of abnormal behaviors. However, most existing behavior recognition models rely on predefined static data sets for training and lack the ability to dynamically adapt to changes in behavior patterns in complex scenarios. This not only limits the generalization ability of the model, but also easily leads to problems such as reduced recognition accuracy and frequent false alarms in actual power grid operation scenarios, affecting the stability and reliability of the system.

[0004] The high variability of power grid operation scenarios requires the behavior recognition model to have the ability to continuously learn and dynamically adapt to new behavior patterns. Especially in long-term deployment and operation, changes in behavior patterns are inevitable. If the model cannot adapt to new data distribution and behavior characteristics in a timely manner, the recognition effect will be significantly reduced. Therefore, it is critical to develop a continuous learning model that can maintain efficient recognition in complex and changeable scenarios. Such models not only need to have dynamic update capabilities in long-term operation, but also should effectively respond to emerging behavior patterns to ensure the long-term security of the system. At the same time, existing behavior recognition models often face a large amount of redundant data during training, resulting in a waste of computing resources and low training efficiency. Especially in power grid operation scenarios, most of the collected data are routine operation behaviors, and it is difficult for the model to effectively distinguish which samples have a key impact on its performance. Therefore, under the framework of continuous learning, how to accurately identify and select key samples that have a significant impact on model training has become a core technical issue for improving model efficiency and adaptability. Summary of the invention

[0005] In view of the problems existing in the prior art, the present invention is proposed.

[0006] Therefore, the problem to be solved by the present invention is how to accurately identify and select key samples that have a significant impact on model training, which becomes a core technical issue in improving model efficiency and adaptability.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0008] In a first aspect, an embodiment of the present invention provides a method for identifying time sequence behavior of power grid operations based on continuous learning, which includes collecting samples containing specific time sequence behaviors in power grid power outage operation scenarios, and selecting key frames and time periods and labeling categories of the samples;

[0009] Extracting and processing features of the sample to obtain a detection sample annotated with a detection frame, and obtaining original spatiotemporal features of the detection sample;

[0010] Build a multi-strategy sample selection module to select challenging samples for model learning and establish a sample memory library;

[0011] Based on the principle of continuous learning, iterative training is performed using samples in the sample memory library to obtain a recognition model;

[0012] The recognition model is used to perform tests and obtain recognition results of specific time sequence behaviors in power grid outage operation scenarios.

[0013] As a preferred solution of the method for identifying time-series behaviors of power grid operations based on continuous learning described in the present invention, the collection of samples containing specific time-series behaviors in power grid power outage operation scenarios includes: video recording with 1080P high-definition resolution, with an average duration of 10 minutes for each video; video data acquisition covers changing conditions of different lighting, time, location and weather; key frames and time periods are selected and categories are labeled by professionals, wherein specific time-series behavior categories include climbing and hanging ground wires.

[0014] As a preferred solution of the method for identifying the timing behavior of power grid operations based on continuous learning described in the present invention, the feature extraction and processing of samples includes: using a target detection model to process the collected video data, automatically detecting and locating the operators in the video, and obtaining detection frame annotation information; using a dual-stream feature extraction network to perform deep feature extraction on the detection results to obtain spatial information and time information, thereby extracting original spatiotemporal features; and performing data enhancement on the original spatiotemporal features to obtain enhanced spatiotemporal features.

[0015] As a preferred solution of the method for identifying the timing behavior of power grid operations based on continuous learning described in the present invention, the construction of a multi-strategy sample selection module includes: inputting the original spatiotemporal features and enhanced spatiotemporal features into the recognition model, performing iterative training, and obtaining the embedded features and errors of each sample; using the errors to calculate the error change rate, and selecting challenging samples according to the threshold; calculating the similarity between samples, and selecting difficult samples with the smallest similarity for training; integrating the selected challenging samples to establish a sample memory library.

[0016] As a preferred solution of the method for identifying the timing behavior of power grid operations based on continuous learning described in the present invention, the method for calculating the error change rate includes: calculating based on the error change value of the sample between consecutive iteration rounds; considering the behavior duration of the sample and the attenuation effect of historical errors; and selecting samples whose error changes exceed the threshold by setting a threshold τ1.

[0017] As a preferred solution of the method for identifying the timing behavior of power grid operation based on continuous learning described in the present invention, wherein: the calculation of the similarity between samples includes: using cosine similarity to calculate the similarity between samples; calculating the expected similarity between the difficult sample and other samples of the same category; calculating the expected change rate of similarity between samples, and according to the threshold τ 2 Filter samples with large change rates.

[0018] As a preferred solution of the method for identifying timing behavior of power grid operations based on continuous learning described in the present invention, the iterative training based on the principle of continuous learning includes: training based on samples in the sample memory library obtained by learning in the first task cycle; training in combination with new samples selected by the multi-strategy sample selection module in the second task cycle; using the cross entropy loss function to optimize the model; dynamically updating the sample memory library for model learning in subsequent task cycles.

[0019] In a second aspect, an embodiment of the present invention provides a power grid operation timing behavior recognition system based on continuous learning, which includes a collection module, which collects samples containing specific timing behaviors in power grid power outage operation scenarios, and selects key frames and time periods and labels categories for the samples;

[0020] An extraction module extracts and processes features of the sample to obtain a detection sample annotated with a detection frame, and simultaneously obtains the original spatiotemporal features of the detection sample;

[0021] Building modules: building a multi-strategy sample selection module, selecting challenging samples for model learning, and building a sample memory library;

[0022] An iterative module, based on the principle of continuous learning, uses the samples in the sample memory library to perform iterative training to obtain a recognition model;

[0023] The identification module uses the identification model to perform testing to obtain identification results of specific time sequence behaviors in the power grid outage operation scenario.

[0024] In a third aspect, an embodiment of the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program instructions are executed by the processor, the steps of the method for identifying timing behavior of power grid operations based on continuous learning as described in the first aspect of the present invention are implemented.

[0025] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program instructions are executed by a processor, the steps of the method for identifying timing behavior of power grid operations based on continuous learning as described in the first aspect of the present invention are implemented.

[0026] The beneficial effects of the present invention are as follows: the present invention forms a comprehensive and real data foundation by using 1080P high-definition resolution to record videos during the data collection stage and covering different lighting, time, location and weather conditions. This diversified data collection method not only improves the representativeness of the samples, but also enhances the model's adaptability to complex environments, solving the problem of unstable recognition effects of traditional methods in complex environments.

[0027] By establishing a dual-stream feature extraction network architecture: the spatial branch processes RGB frame information; the temporal branch processes optical flow information; the collaborative extraction of spatiotemporal features is achieved. Compared with the single feature extraction method, the ability to capture dynamic behaviors is improved, and the recognition accuracy is increased by about 15%.

[0028] By innovatively constructing a multi-strategy sample selection module: introducing an error change rate calculation mechanism, taking into account the sample behavior duration and historical error attenuation;

[0029] Cosine similarity is used to measure the similarity between samples; this enables the model to automatically screen challenging key samples, greatly reducing the redundancy of training data (by about 40%) while maintaining model performance and significantly improving training efficiency.

[0030] By designing a sample memory library storage mechanism: dynamically updating the storage strategy; combining the learning process of two task cycles; and effectively solving the problem that traditional models easily forget learned experience, the recognition accuracy in new scenarios is improved by 1.91% (from 82.78% to 84.69%) compared with traditional methods.

[0031] The overall synergistic effect of the system: a complete technical chain is formed from data collection to feature extraction, and then to model training; accurate identification of two key operational behaviors, climbing and hanging ground wires, is achieved; the accuracy rate of climbing action recognition reaches 85.71%; the accuracy rate of hanging ground wire action recognition reaches 80.49%; it provides reliable technical support for the safe monitoring of power grid operations.

[0032] Practical application value: It realizes real-time behavior monitoring in complex environments; supports continuous optimization and updating of models; reduces manual monitoring costs; improves operation safety; and is particularly suitable for application in high-risk operation scenarios such as power grid construction and maintenance.

[0033] By taking into account both the change in sample error and the change in expected similarity, we have achieved accurate positioning of difficult samples, an effect that exceeds the expectations of traditional single evaluation indicators. Without increasing hardware costs, we have significantly improved system performance through algorithm optimization. The model has the ability to continuously learn and can constantly adapt to new operating scenarios, which breaks through the limitations of traditional static models. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0035] Figure 1 A flowchart of a method for identifying timing behavior of power grid operations based on continuous learning;

[0036] Figure 2 A computer equipment diagram for a method for identifying the timing behavior of power grid operations based on continuous learning;

[0037] Figure 3 The original video frame image in the power grid outage operation scene of the power grid operation time sequence behavior recognition method based on continuous learning;

[0038] Figure 4 This is an example diagram of the recognition results in a power grid outage operation scenario using a power grid operation timing behavior recognition method based on continuous learning. DETAILED DESCRIPTION

[0039] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0040] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0041] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor is it a separate or selective embodiment that is mutually exclusive with other embodiments.

[0042] Example 1

[0043] Reference Figure 1-2 , which is the first embodiment of the present invention, and provides a method for identifying timing behavior of power grid operations based on continuous learning, comprising:

[0044] S100: Collect samples containing specific time series behaviors in power grid outage operation scenarios, select key frames and time periods, and label categories for the samples;

[0045] It is worth noting that in the preferred embodiment of the present invention, collecting videos of power grid power outage operation scenes for annotation is a non-restrictive and preferred choice, and those skilled in the art can collect sample data in different power grid scenarios for annotation according to identification requirements.

[0046] S101: Collect samples containing specific temporal behaviors in power grid outage operation scenarios, including: video recording with 1080P high-definition resolution, with an average duration of 10 minutes per video; video data collection covers changing conditions of different lighting, time, location and weather; key frames and time periods are selected and categorized by professionals, among which specific temporal behavior categories include climbing and hanging ground wires.

[0047] S200: extracting and processing features of the sample to obtain a detection sample annotated with a detection frame, and obtaining the original spatiotemporal features of the detection sample;

[0048] Preferably, based on the existing detection model, a detection sample containing a detection box annotation is obtained, and at the same time, the original spatiotemporal features of the detection sample are obtained using the existing two-stream feature extraction network, and the original spatiotemporal features are data enhanced to obtain enhanced spatiotemporal features.

[0049] S201: Feature extraction and processing of samples include: using the target detection model to process the collected video data, automatically detecting and locating the workers in the video, and obtaining detection frame annotation information; using the dual-stream feature extraction network to perform deep feature extraction on the detection results to obtain spatial information and time information, thereby extracting original spatiotemporal features; performing data enhancement on the original spatiotemporal features to obtain enhanced spatiotemporal features.

[0050] S300: Build a multi-strategy sample selection module to select challenging samples for model learning and establish a sample memory library;

[0051] Preferably, a multi-strategy sample selection module is constructed to consider the variation of sample errors and the expected variation of similarity between samples, select challenging samples, conduct model learning for the first task cycle, and establish a sample memory library to store these samples.

[0052] S301: Constructing a multi-strategy sample selection module includes: inputting the original spatiotemporal features and enhanced spatiotemporal features into the recognition model, performing iterative training, and obtaining the embedded features and errors of each sample; using the error to calculate the error change rate, and selecting challenging samples according to the threshold; calculating the similarity between samples, and selecting the difficult samples with the smallest similarity for training; integrating the selected challenging samples and establishing a sample memory library.

[0053] S302: The method for calculating the error change rate includes: calculating based on the error change value of the sample between consecutive iteration rounds; considering the behavior duration of the sample and the attenuation effect of the historical error; and selecting the sample whose error change exceeds the threshold by setting a threshold τ1.

[0054] S400: Based on the principle of continuous learning, iterative training is performed using samples in the sample memory library to obtain a recognition model;

[0055] Preferably, based on the principle of continuous learning, the recognition model is iteratively trained again using samples in the sample memory library and samples selected by the multi-strategy sample selection module in the second task cycle learning to obtain the best recognition model.

[0056] S401: Iterative training based on the principle of continuous learning includes: training based on samples in the sample memory library obtained through learning in the first task cycle; training with new samples selected by the multi-strategy sample selection module in the second task cycle; optimizing the model using the cross entropy loss function; and dynamically updating the sample memory library for model learning in subsequent task cycles.

[0057] S500: Performing a test using the recognition model to obtain recognition results of specific time sequence behaviors in a power grid outage operation scenario.

[0058] Furthermore, this embodiment also provides a power grid operation timing behavior recognition system based on continuous learning, comprising:

[0059] The collection module collects samples containing specific time series behaviors in power grid outage operation scenarios, selects key frames and time periods, and labels categories for the samples;

[0060] The extraction module extracts and processes features of the samples to obtain detection samples annotated with detection boxes and the original spatiotemporal features of the detection samples.

[0061] Building modules: building a multi-strategy sample selection module, selecting challenging samples for model learning, and building a sample memory library;

[0062] The iterative module, based on the principle of continuous learning, uses the samples in the sample memory library for iterative training to obtain the recognition model;

[0063] The recognition module uses the recognition model to perform tests and obtains the recognition results of specific timing behaviors in the power grid outage operation scenario.

[0064] In summary, by using 1080P high-definition resolution video recording during the data collection phase and covering different lighting, time, location and weather conditions, a comprehensive and real data foundation is formed. This diversified data collection method not only improves the representativeness of the samples, but also enhances the model's ability to adapt to complex environments, solving the problem of unstable recognition effects of traditional methods in complex environments.

[0065] By establishing a dual-stream feature extraction network architecture: the spatial branch processes RGB frame information; the temporal branch processes optical flow information; the collaborative extraction of spatiotemporal features is achieved. Compared with the single feature extraction method, the ability to capture dynamic behaviors is improved, and the recognition accuracy is increased by about 15%.

[0066] By innovatively constructing a multi-strategy sample selection module: introducing an error change rate calculation mechanism, taking into account the sample behavior duration and historical error attenuation;

[0067] Cosine similarity is used to measure the similarity between samples; this enables the model to automatically screen challenging key samples, greatly reducing the redundancy of training data (by about 40%) while maintaining model performance and significantly improving training efficiency.

[0068] By designing a sample memory library storage mechanism: dynamically updating the storage strategy; combining the learning process of two task cycles; and effectively solving the problem that traditional models easily forget learned experience, the recognition accuracy in new scenarios is improved by 1.91% (from 82.78% to 84.69%) compared with traditional methods.

[0069] The overall synergistic effect of the system: a complete technical chain is formed from data collection to feature extraction, and then to model training; accurate identification of two key operational behaviors, climbing and hanging ground wires, is achieved; the accuracy rate of climbing action recognition reaches 85.71%; the accuracy rate of hanging ground wire action recognition reaches 80.49%; it provides reliable technical support for the safe monitoring of power grid operations.

[0070] Practical application value: It realizes real-time behavior monitoring in complex environments; supports continuous optimization and updating of models; reduces manual monitoring costs; improves operation safety; and is particularly suitable for application in high-risk operation scenarios such as power grid construction and maintenance.

[0071] By taking into account both the change in sample error and the change in expected similarity, we have achieved accurate positioning of difficult samples, an effect that exceeds the expectations of traditional single evaluation indicators. Without increasing hardware costs, we have significantly improved system performance through algorithm optimization. The model has the ability to continuously learn and can constantly adapt to new operating scenarios, which breaks through the limitations of traditional static models.

[0072] Example 2

[0073] Reference Figure 2 - Figure 4 , which is the second embodiment of the present invention, and this embodiment provides a method for identifying the timing behavior of power grid operations based on continuous learning. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0074] In the power grid outage operation scenario, data collection is carried out by combining on-site video recording with a real-time monitoring system. Video recording uses 1080P high-definition resolution, and the average duration of each video is about 10 minutes, ensuring that the entire operation process is recorded. Video data collection covers a variety of complex environments, including changing conditions such as lighting, time, location and weather. For example, data is recorded at different time periods within a day to capture scenes from bright sunlight to low-light environments; collected at different work sites to ensure location diversity; and covers a variety of weather conditions such as sunny, cloudy and rainy days to simulate the variability of actual work scenes. All collected video data is selected by professionals for key frames and time periods, as well as category annotation. Specific temporal behavior categories include climbing and hanging ground wires.

[0075] In the preferred embodiment of the present invention, the collected data includes 870 videos and 2107 action examples, including 760 pole climbing and 313 grounding wire hanging, and the action data retention rate is about 1 / 6. The training set and the validation set are divided into about 3:1, wherein the training set includes 642 videos, including 592 pole climbing actions and 272 grounding wire hanging actions; the validation set includes 228 videos, including 168 pole climbing actions and 41 grounding wire hanging actions.

[0076] Based on the existing detection model, a detection sample containing a detection box annotation is obtained. At the same time, the original spatiotemporal features of the detection sample are obtained using the existing two-stream feature extraction network, and the original spatiotemporal features are data enhanced to obtain enhanced spatiotemporal features.

[0077] Preferably, based on the existing target detection model (yolov5), the collected video data is processed to automatically detect and locate the workers in the video. The detection model accurately selects the area of ​​the workers in each frame by generating detection frame annotation information. To ensure the reliability of the detection, the model will be verified based on multiple frames of continuous images, thereby effectively reducing missed detections and false detections, especially in work scenes under complex backgrounds and lighting conditions. The sample data after detection contains detailed worker location and boundary information;

[0078] Based on the detection results, the existing two-stream feature extraction network (I3D) is further used to perform deep feature extraction on the detection results. The two-stream network includes two parallel branches, one for processing spatial information (RGB frames) and the other for processing temporal information (optical flow), so as to extract the original spatiotemporal features in the collected samples and perform data enhancement (such as temporal masking, temporal shifting, scaling, random noise, random abstention, etc.) on the original spatiotemporal features to obtain enhanced spatiotemporal features.

[0079] Construct a multi-strategy sample selection module, consider the change of sample error and the expected change of similarity between samples, select challenging samples, conduct model learning in the first task cycle, and establish a sample memory library to store these samples.

[0080] Preferably, the original spatiotemporal features and the enhanced spatiotemporal features are input into the recognition model, and the model is iteratively trained to obtain the embedded features of each sample and the error between the predicted result and the true result of each sample, which are input into the multi-strategy sample selection module;

[0081] The error change rate is calculated using the error, the input samples are sorted according to the calculated error change rate, and challenging samples are selected according to the threshold;

[0082] Use the sample selected in the previous step to obtain the difficult sample with the minimum similarity. According to the true category label, use the difficult sample and other samples in its category to calculate the expected change rate of similarity between samples, and filter samples with large change rates according to the threshold.

[0083] Integrate the challenging samples selected in the above two steps, build a sample memory library, and perform iterative training of the recognition model in the first task cycle.

[0084] Preferably, in order to meet the requirements of continuous learning, it is assumed that two task cycles are learned. During the model training process of the first task cycle, the recognition error of each sample is calculated as follows:

[0085]

[0086] Where N is the number of samples, y is the true result, is the prediction result. By evaluating the loss of the training samples, the error value of each sample is obtained. And record the error.

[0087] In a preferred embodiment of the present invention, N=4.

[0088] Preferably, based on the recorded error, the error change rate of the sample between two consecutive iterations t and t+1 is defined, which is calculated as follows:

[0089]

[0090] in, It represents the error change value of the sample between the tth round and the t+1th round, which is the basis of the error change rate. i Represents sample x i The duration of the behavior, longer behaviors are usually of higher complexity. It is a decay function of historical error. As the error of the sample in history decreases, its influence on the error change rate gradually decreases. Here, λ is the decay coefficient, which controls the influence of historical error. |·| represents the absolute value. α and γ adjust the influence ratio of the basic error change term and the historical error term respectively. Through these two parameters, their respective contributions to the error change rate can be balanced.

[0091] In order to determine the samples with significant error changes, a threshold τ is set. 1 , select the error change Δe i Samples exceeding the threshold in the following way:

[0092] F={x i |Δe i >τ 1}i=1,2,N.

[0093] in, is the selected sample set, B is the number of selected samples. These samples often represent the parts that have not been well mastered in model learning and have higher training value.

[0094] In the preferred embodiment of the present invention, α=0.5, γ=0.5, λ=0.5, τ 1 =0.4.

[0095] Preferably, the selected sample set F, for each sample x in F i , calculate its difference with all other samples x in the current category j The similarity S(x i ,x j ), and select the sample with the smallest similarity as the difficult sample. The similarity can be calculated using the cosine similarity formula:

[0096]

[0097] Among them, v i and v j Respectively represent samples x i and sample x j The embedding features of , ||·|| represents the 2-norm. By calculation, we get i The hard sample x with the smallest similarity d ,Right now:

[0098]

[0099] According to the true category label, extract all samples in the category and calculate the difficult sample x d Compared with other samples x of the same category k The similarity expectation is as follows:

[0100]

[0101] Where c represents the cth class, and |c| is the number of samples in the cth class. The expected change rate of similarity between samples between two consecutive iterations t and t+1 is defined as follows:

[0102]

[0103] Set the threshold τ of the expected change rate of similarity 2 , filter out the sample set whose change rate is greater than the threshold H is the number of selected samples.

[0104] In a preferred embodiment of the present invention, τ 2 =0.6.

[0105] Preferably, the sample set F of the step (using the error to calculate the error change rate, sorting the input samples according to the calculated error change rate, and selecting challenging samples according to the threshold) and the sample set G of the step (using the sample selected in the previous step to obtain the difficult sample with the minimum similarity, using the difficult sample and other samples in its category according to the true category label to calculate the expected change rate of similarity between samples, and screening samples with a large change rate according to the threshold) are selected, and the two are integrated to construct a sample memory library The cross entropy loss is used for iterative training of the recognition model in the following form:

[0106]

[0107] Among them, y k is the true value label of the kth sample, is the prediction result of the recognition model for the kth sample.

[0108] In a preferred embodiment of the present invention, AdamW optimizer is used as the model optimizer, the initial value of the learning rate is set to 0.0001, the learning rate decay adopts cosine annealing decay, and a total of 40 cycles are trained.

[0109] It is worth noting that in the preferred embodiment of the present invention, the network training method and the parameter configuration of the learning rate optimization strategy are a non-restrictive and preferred choice. Those skilled in the art can select the model training method and parameter configuration based on various indicators such as recognition accuracy and efficiency.

[0110] Based on the principle of continuous learning, the recognition model is iteratively trained again using the samples in the sample memory library and the samples selected by the multi-strategy sample selection module in the second task cycle learning to obtain the optimal recognition model.

[0111] Preferably, the recognition model is iteratively trained again based on the sample memory library obtained by learning in the first task cycle, using the samples stored therein and the samples selected by the multi-strategy sample selection module in the second task cycle learning, to obtain the best recognition model;

[0112] The samples selected by the multi-strategy sample selection module in the second task cycle learning are stored or dynamically updated to the sample memory library to help the model learning of the next task cycle.

[0113] In a preferred embodiment of the present invention, assuming that two task cycles are learned, the first task cycle randomly selects 40% of the samples from the training set as input for recognition model learning, and the remaining samples are left for model learning in the second task cycle.

[0114] The optimal recognition model was used for testing and the recognition results of specific timing behaviors in power grid outage operation scenarios were obtained.

[0115] Preferably, the trained optimal recognition model is deployed on the test system, and the newly collected sample data in the power grid outage operation scenario is used as input to perform the test and obtain the recognition result.

[0116] For the behavior recognition model obtained in the preferred embodiment of the present invention, 150 sample data newly collected in the power grid power outage operation scenario were used to perform 10 tests, and the average accuracy of sample recognition was recorded. The results are shown in Table 1.

[0117] In a preferred embodiment of the present invention, the trained model is deployed on a test system, and the test samples are used to obtain the recognition results of the model's specific timing behaviors in the power grid outage operation scenario. The software configuration of the test system is python3.9.8, pytorch2.0.1, cuda12.5, and the hardware configuration is an L20 graphics card (48G video memory).

[0118] Some test results are shown below: Figure 4 As shown in the test result diagram, it can be seen that the behavior recognition model of this embodiment can accurately identify specific time series behaviors in the test sample data after continuous learning of samples in the power grid power outage operation scenario.

[0119] As can be seen from Table 1, the behavior recognition model of this embodiment is tested on newly collected data in the power grid outage operation scenario, and the accuracy of climbing and hanging ground wires in this method is 85.71% and 80.49%, and the average accuracy is 84.69%, which is 1.91% higher than the average accuracy of the baseline method of 82.78%, where the baseline method is the result of no sample selection and storage. The results of this method show that the method improves the accuracy of the specific time series behavior recognition model by continuously learning new samples.

[0120] Table 1 Comparison of model results on test data

[0121] Behavior Category Baseline method (accuracy) This method (accuracy) Climb 84.52％ 85.71％ Hanging ground wire 75.61％ 80.49％

[0122] The beneficial effect of the present invention is that, compared with the prior art, it aims to develop an effective sample selection strategy based on a continuous learning framework, which constructs a multi-strategy sample selection module, focuses on sample error changes and expected changes in similarity between samples, and dynamically selects key samples with high learning value for the model. On this basis, combined with continuous learning technology, the model is iteratively optimized to establish a sample memory library with adaptive capabilities to reduce the forgetting effect and improve the accuracy of identifying specific temporal behaviors in power grid outage operation scenarios. This method continuously updates and optimizes the model to ensure that it always maintains efficient and reliable recognition performance in the face of the diversity and complexity of the power grid outage operation environment, thereby providing a strong technical guarantee for the safe and stable operation of the power system.

[0123] Example 3

[0124] This embodiment also provides a computer device, which is suitable for a method for identifying the timing behavior of power grid operations based on continuous learning, and includes a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement a forced oscillation detection and positioning method for a distribution network as proposed in the above embodiment.

[0125] This embodiment further provides a storage medium on which a computer program is stored. When the program is executed by a processor, a forced oscillation detection and positioning method for a distribution network is implemented as proposed in the above embodiment.

[0126] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0127] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0128] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0129] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering, or processing in another suitable manner as necessary, and then stored in a computer memory.

[0130] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used to implement: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0131] 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 preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for identifying timing behavior of power grid operations based on continuous learning, characterized in that: include, Collect samples containing specific time series behaviors in power grid outage operation scenarios, and select key frames and time periods and label categories for the samples; Extracting and processing features of the sample to obtain a detection sample annotated with a detection frame, and obtaining original spatiotemporal features of the detection sample; Build a multi-strategy sample selection module to select challenging samples for model learning and establish a sample memory library; Based on the principle of continuous learning, iterative training is performed using samples in the sample memory library to obtain a recognition model; The recognition model is used to perform tests and obtain recognition results of specific time sequence behaviors in power grid outage operation scenarios.

2. The method for identifying power grid operation timing behavior based on continuous learning according to claim 1, characterized in that: The samples collected in the power grid outage operation scenario containing specific temporal behaviors include: video recording with 1080P high-definition resolution, with each video having an average duration of 10 minutes; video data collection covers changing conditions of different lighting, time, location and weather; key frames and time periods are selected and categorized by professionals, where specific temporal behavior categories include climbing and hanging ground wires.

3. The method for identifying power grid operation timing behavior based on continuous learning according to claim 2, characterized in that: The feature extraction and processing of samples includes: using a target detection model to process the collected video data, automatically detecting and locating the workers in the video, and obtaining detection frame annotation information; using a dual-stream feature extraction network to perform deep feature extraction on the detection results to obtain spatial information and time information, thereby extracting original spatiotemporal features; and performing data enhancement on the original spatiotemporal features to obtain enhanced spatiotemporal features.

4. The method for identifying power grid operation timing behavior based on continuous learning according to claim 3, characterized in that: The construction of the multi-strategy sample selection module includes: inputting the original spatiotemporal features and the enhanced spatiotemporal features into the recognition model, performing iterative training, and obtaining the embedded features and errors of each sample; using the errors to calculate the error change rate, and selecting challenging samples according to a threshold; calculating the similarity between samples, and selecting difficult samples with the smallest similarity for training; integrating the selected challenging samples and establishing a sample memory library.

5. The method for identifying power grid operation timing behavior based on continuous learning according to claim 4, characterized in that: The method for calculating the error change rate includes: calculating based on the error change value of the sample between consecutive iteration rounds; considering the behavior duration of the sample and the attenuation effect of the historical error; and selecting the sample whose error change exceeds the threshold by setting a threshold τ1.

6. The method for identifying power grid operation timing behavior based on continuous learning according to claim 5, characterized in that: The calculation of similarity between samples includes: using cosine similarity to calculate the similarity between samples; calculating the expected similarity between difficult samples and other samples of the same category; calculating the expected change rate of similarity between samples, and screening samples with large change rates according to a threshold τ2.

7. The method for identifying power grid operation timing behavior based on continuous learning according to claim 6, characterized in that: The iterative training based on the principle of continuous learning includes: training based on samples in the sample memory library obtained by learning in the first task cycle; training in combination with new samples selected by the multi-strategy sample selection module in the second task cycle; using the cross entropy loss function to optimize the model; and dynamically updating the sample memory library for model learning in subsequent task cycles.

8. A system for identifying the timing behavior of power grid operations based on continuous learning, based on the method for identifying the timing behavior of power grid operations based on continuous learning according to any one of claims 1 to 7, characterized in that: Also includes, A collection module collects samples containing specific time series behaviors in power grid outage operation scenarios, and selects key frames and time periods and labels categories for the samples; An extraction module extracts and processes features of the sample to obtain a detection sample annotated with a detection frame, and simultaneously obtains the original spatiotemporal features of the detection sample; Building modules: building a multi-strategy sample selection module, selecting challenging samples for model learning, and building a sample memory library; An iterative module, based on the principle of continuous learning, uses the samples in the sample memory library to perform iterative training to obtain a recognition model; The identification module uses the identification model to perform testing to obtain identification results of specific time sequence behaviors in the power grid outage operation scenario.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method for identifying the timing behavior of power grid operations based on continuous learning are implemented in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for identifying the timing behavior of power grid operations based on continuous learning according to any one of claims 1 to 7 are implemented.

Images