Load switch control method of distribution box

Abstract

The invention discloses a load switch control method of a distribution box, and relates to the technical field of automatic control systems, and the method comprises the following steps: obtaining real-time load data of a to-be-detected region, building a dynamic switch threshold control model based on LSTM, and generating an initial threshold interval; constructing a load behavior map according to the optimal clustering model, mapping the current load data to the load behavior map, and regulating and controlling the initial threshold interval in combination with a historical load data fluctuation interval to obtain a corrected threshold interval; extracting a distribution feature vector and a classification precision parameter of the current load data as first data, and training an optimal classification model based on the first data; inputting the corrected threshold interval into the optimal classification model, and identifying a current load state label; according to the method, the load behavior graph is constructed, the load state is evaluated in combination with the fluctuation interval of the historical load data, then the control strategy is adjusted, dynamic regulation and control of the switch threshold interval are achieved, and the load switch is controlled according to the correction result.

Description

Technical Field

[0001] The present invention relates to the technical field of automatic control systems, and more specifically, to a method for controlling a load switch of a distribution box. Background Art

[0002] In the context of the rapid development of modern smart grids, the distribution system, as a key link connecting the power generation side and the user side, has increasingly attracted attention for its operational safety, stability, and economy. Especially in scenarios such as industrial parks, residential communities, and commercial complexes where the electrical load exhibits high-frequency fluctuations and complex dynamic characteristics, traditional load switch control methods have become difficult to meet the high standards for energy efficiency optimization and safety management. Currently, most control strategies for distribution box load switches still rely on fixed thresholds or simple rules, lacking the ability to perceive and respond to real-time load change trends, and are prone to problems such as mis-triggering, response lag, or energy consumption waste, severely restricting the improvement of the regulation efficiency and automation level of the distribution system. Therefore, in response to the high-precision regulation requirements of future smart distribution systems, a load switch control method with adaptive learning and dynamic adjustment capabilities is needed to achieve more refined and intelligent load management.

[0003] For example, an intelligent switch device adaptive control method and its system disclosed in the invention patent with the publication number CN119105293B relate to the technical field of intelligent switch device control. The intelligent switch device adaptive control method includes the following steps: load data acquisition and processing, fluctuation compliance evaluation and judgment, balance optimization evaluation and judgment, and fault detection evaluation and feedback. By preprocessing the initial load data to obtain load-related data, then evaluating the load in a preset industrial park area based on the load-related data to determine whether to perform load balance optimization, then evaluating the effect of load balance optimization and determining whether to perform device fault detection, and finally obtaining the switch device fault evaluation index after performing device fault detection and conducting a comprehensive evaluation to obtain a comprehensive evaluation result, the effect of improving the adaptive control efficiency of the intelligent switch device is achieved, and the problem of low adaptive control efficiency of intelligent switch devices in the prior art is solved.

[0004] For example, an optimization method, system, device, and medium for a flexible control device of electric power load disclosed in the invention patent with the publication number of CN118466219B. By obtaining the historical electric power load data of the target power grid, a historical electric power load database is constructed; based on the historical electric power load database, a first partial optimization result of the flexible control device of electric power load is obtained; the real-time operation data of the target power grid is obtained to update the historical electric power load database, and an optimized electric power load database is obtained; based on the optimized electric power load database, the first partial optimization result is corrected to obtain a second partial optimization result, and the flexible control device of electric power load of the target power grid is globally optimized based on the second partial optimization result; the flexible control device of electric power load of the target power grid is dynamically optimized based on the global optimization result. The method provided by this application realizes the local optimization, global optimization, and dynamic optimization of the flexible control device of electric power load, improves the control effect, and meets the requirements of power grid intelligence.

[0005] In the above disclosed technical solution, there are at least the following technical problems: The traditional load control method of distribution boxes is mostly based on static threshold setting, and cannot automatically adjust the control strategy according to the real-time fluctuation of the load, resulting in a slow response to sudden load abnormalities and affecting the system stability. At the same time, the traditional load state recognition method fails to adaptively optimize the classification features according to different data distribution characteristics, and the lack of a fusion mechanism limits the recognition accuracy. In view of the above problems, the present invention proposes a solution. Summary of the Invention

[0006] In order to overcome the above defects of the prior art, an embodiment of the present invention provides a method for controlling a load switch of a distribution box. By constructing a load behavior map and combining the fluctuation range of historical load data, the load state can be evaluated more scientifically, and then the control strategy can be adjusted to achieve the dynamic regulation of the switch threshold range, and the load switch is controlled according to the correction result.

[0007] To achieve the above object, the present invention provides the following technical solutions: A method for controlling a load switch of a distribution box includes the following steps: obtaining the real-time load data of the area to be measured, and constructing a dynamic switch threshold control model based on LSTM to generate an initial threshold range; constructing a load behavior map according to the optimal clustering model, mapping the current load data to the load behavior map, and regulating the initial threshold range in combination with the historical load data fluctuation range to obtain a corrected threshold range; extracting the distribution feature vector and classification accuracy parameter of the current load data as the first data, and training the optimal classification model based on the first data; inputting the corrected threshold range into the optimal classification model to identify the current load state label.

[0008] In a preferred embodiment, the real-time load data of the area to be measured is obtained, and a dynamic switch threshold control model is constructed based on LSTM to generate an initial threshold range, which is specifically as follows: The phase lines and circuits of the distribution box in the area to be measured are monitored in real time, and the electrical parameter data of each circuit is synchronously collected at a fixed period to form multi-channel parallel load data; the real-time load data is preprocessed by an edge computing node, and the time synchronization of each data stream is performed based on the NTP protocol; the preprocessed load data is organized in chronological order, and the sliding window technology is used for time series recombination, and the load data of every N consecutive time points constitutes a single time series sample, and each sample constitutes a dimensional tensor; the time series sample is input into the LSTM model for training to obtain a trained dynamic switch threshold model, and the load fluctuation trend is predicted based on the LSTM model; according to the predicted value output by the LSTM model and combined with the dynamic switch threshold model, a dynamic threshold range is constructed.

[0009] In a preferred embodiment, the dynamic threshold range is constructed according to the predicted value output by the LSTM model and combined with the dynamic switch threshold model, which is specifically as follows: Based on the load data within the window, the standard deviation of the load data within the window is output; according to the predicted load trend and the standard deviation of the load data within the sliding window, a dynamic threshold model is constructed to obtain a dynamic threshold range.

[0010] In a preferred embodiment, the optimal clustering model is specifically as follows: The load data is preprocessed, and the load feature data is dimensionally reduced based on the principal component analysis method to obtain sample data; based on the locality-sensitive hashing algorithm, similar sample data is mapped to the same bucket through hash encoding to generate hash values; the hash values are clustered, and based on the cosine similarity method, the local neighborhood of each data point is obtained; according to the data in the local neighborhood, local features are extracted; according to the local features, the clustering effect of the clustering model is evaluated, and the best clustering model is selected.

[0011] In a preferred embodiment, the load behavior map is constructed according to the optimal clustering model, which is specifically as follows: The historical load data of the area to be measured is obtained and segmented by a sliding window, and each window contains a time series sample of a preset length; the time series samples are constituted into tensors, and the tensors are input into the clustering model for training, and the load behavior category to which each time window belongs is output; the sample tensors are clustered by the trained clustering model, and according to the clustering result, the time window data belonging to the same category is grouped into the same group to obtain several load behavior categories, and each category constitutes a load behavior map.

[0012] In a preferred embodiment, mapping the current load data to the load behavior atlas and adjusting the initial threshold range in combination with the historical load data fluctuation range are specifically as follows: extracting the features of the real-time load data within the sliding window at the current moment to form a tensor with the same dimension as the load behavior atlas; obtaining the similarity between the current feature vector and the central vectors of all load behavior atlases based on the cosine similarity method, and selecting the load behavior category atlas with the highest similarity as the mapping atlas for the current window; extracting the maximum deviation values within the historical load fluctuation range of this category from the mapping atlas, which are respectively used to represent the upper offset correction amount and the lower offset correction amount as correction factors; controlling the initial threshold range through the correction factors in combination with the standard deviation of the sliding window load data to obtain the corrected threshold range.

[0013] In a preferred embodiment, extracting the distribution feature vector and classification accuracy parameter of the current load data as the first data and training the optimal classification model based on the first data are specifically as follows: obtaining the first data of the load data, where the first data includes distribution features and classification accuracy; obtaining the load data of the area to be measured, obtaining the distribution features of the load data, and analyzing the distribution pattern of the load data through the histogram method; screening the load data features based on the distribution features of the load data and constructing a load data feature subset; inputting the load data feature subset into the classification model as a time series sample for classification, and evaluating the classification accuracy of the classification model based on cross-validation, and screening out the best classification model according to the classification accuracy.

[0014] In a preferred embodiment, screening the load data features based on the distribution features of the load data is specifically as follows: obtaining the distribution pattern of the load data and extracting the feature indicators related to the data distribution pattern; evaluating the contribution degree of each feature indicator to the load data classification performance based on the feature importance ranking method of correlation analysis; according to the evaluation results, screening out the feature indicators that have the best effect on improving the classification accuracy under the current load data distribution features and constructing a load data feature subset.

[0015] In a preferred embodiment, inputting the corrected threshold range into the optimal classification model to identify the current load status label is specifically as follows: clustering each data point through the best clustering model and outputting the initial load status label; inputting the load feature data of the current window into the trained classification model to obtain the predicted load status label; fusing the output results of the clustering model and the classification model; using the fused load status label as the final identification result and controlling the load switch of the target distribution box.

[0016] The technical effects and advantages of a load switch control method for a distribution box according to the present invention: 1. The present invention introduces a long short-term memory network (LSTM) to perform time series analysis on the load data of the distribution box, thereby realizing the construction of a dynamic threshold model. The LSTM model has a strong ability to capture time series dependencies and can effectively learn the historical fluctuation pattern of load data, thereby predicting the load fluctuation trend. By combining the prediction results with the load standard deviation in the sliding window, the load control threshold is dynamically adjusted so that the load switch can automatically adjust according to the real-time load conditions. The biggest advantage of this method is that it can adaptively respond to scenarios with large load fluctuations, avoiding the defects of the traditional fixed threshold method, and can more accurately identify the real-time status of the load.

[0017] 2. The present invention performs sliding window segmentation on historical load data and generates a load behavior map using a clustering algorithm. The present invention can quickly identify the current load state under different load fluctuation modes. The construction of the load behavior map analyzes typical load fluctuation modes, such as peak load, valley load, and periodic load, so that the distribution system can map the load state according to the similarity between the real-time data window and the load behavior map. Through this method, the behavior category corresponding to the current load data can be accurately identified, and the control strategy can be adjusted in a targeted manner, effectively improving the prediction accuracy of the load fluctuation range and the control response speed.

[0018] 3. The present invention uses a variety of clustering models (such as clustering methods based on local sensitive hashing) to perform pattern recognition on historical load data, construct load behavior maps, and provide data basis for different load states. After combining with the LSTM prediction model, customized threshold adjustments can be made under different load behavior categories, reflecting the system's adaptability and scalability to complex environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 The present invention is a flow chart of a load switch control method for a distribution box. DETAILED DESCRIPTION

[0020] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0021] Embodiment 1, Figure 1 The present invention provides a load switch control method for a distribution box, comprising the following steps: S1, obtain historical load data, and adjust the pre-acquired dynamic switch threshold according to the fluctuation range of the historical load data.

[0022] In this embodiment, historical load data is obtained, and the previously obtained dynamic switching threshold is adjusted according to the fluctuation range of the historical load data, specifically as follows: Configure electrical parameter acquisition devices such as voltage, current, and frequency in the distribution box system, and monitor each phase line and loop in the area to be measured in real time. Synchronously collect the active power, reactive power, effective voltage value, effective current value, and frequency parameters of each loop at a fixed period to form multi-channel parallel load data; The edge computing node performs millisecond-level preprocessing on the real-time load data, and synchronizes the time of each sensor data stream based on the NTP protocol to generate a structured data packet with a global timestamp; The preprocessed load data is organized in chronological order, and the sliding window technology is used for time series recombination. The load data at every N consecutive time points (for example, N = 30) constitutes a time series sample, and each sample forms a tensor with a dimension of where F is the feature dimension, and the data at each time point contains several load characteristics (such as active power, reactive power, voltage, current, and frequency, etc.); The time series sample is input into the long short-term memory neural network (LSTM) model for training to obtain a trained dynamic switching threshold model. The LSTM model learns the temporal dependence relationship in the historical load data to predict the load fluctuation trend at the next moment; According to the prediction value output by the LSTM model and combined with the dynamic switching threshold model, a dynamic threshold range is constructed.

[0023] In this implementation, according to the prediction value output by the LSTM model and combined with the dynamic switching threshold model, a dynamic threshold range is constructed, specifically as follows: Based on the load data within the window, the standard deviation of the load data within the window is output to measure the fluctuation intensity of the load data; According to the predicted load trend and the standard deviation of the load data within the sliding window, a dynamic threshold model is constructed, and a dynamic threshold range is obtained according to the dynamic threshold model.

[0024] The dynamic threshold model is specifically as follows:

[0025]

[0026] In the formula: is the load upper limit, is the predicted load value at the next moment, is the load upper limit, is a predetermined constant coefficient, and this coefficient is adjusted according to the tolerance of load fluctuations, is the standard deviation of the load data within the sliding window.

[0027] It should be noted that the adaptive adjustment function of the dynamic threshold self-optimizes according to the real-time fluctuations of the load data. When the standard deviation of the load data is large, the threshold range expands to adapt to large fluctuations; when the load data fluctuates little, the threshold range narrows to reduce the risk of false alarms. In addition, a smoothing factor can be introduced in the adjustment of the dynamic threshold to avoid overly drastic threshold changes caused by abnormal fluctuations in a short period of time. The smoothing factor can be implemented by the exponential smoothing method to ensure the stability of the threshold.

[0028] S2. Obtain the clustering effects of different clustering models and obtain the optimal clustering model.

[0029] In this embodiment, obtaining the clustering effects of different clustering models and obtaining the optimal clustering model are specifically as follows: Preprocess the load characteristic data, and reduce the dimension of the load characteristic data based on the principal component analysis method to obtain sample data; Based on the locally sensitive hashing algorithm, map similar sample data to the same bucket through hash coding to generate hash values; Cluster the hash values, and based on the cosine similarity method, obtain the local neighborhood of each data point; Extract local features according to the data in the local neighborhood. The local features include clustering spacing, clustering boundary clarity, silhouette coefficient, cohesion, and separation; Evaluate the clustering effects of the clustering models according to the local features, and screen out the best clustering model according to the evaluation results. The clustering models include K-means and hierarchical clustering.

[0030] The specific formula for cohesion is as follows:

[0031] The specific formula for separation is as follows:

[0032] The specific formula for evaluating the clustering effect is as follows:

[0033] In the formula: is cohesion, is the k-th cluster, is the i-th data point within the cluster, is the j-th data point within the cluster, is separation, is the cluster center of the k-th cluster, is the cluster center of the m-th cluster, is the number of samples in the dataset, that is, the total number of all data points, is the evaluation result of the clustering effect, is the total number of clusters.

[0034] It should be noted that a high cohesion degree means that the data points within the same cluster are very close or tight. This means that the data points within the cluster have a high similarity, and the clustering result is more accurate and effective. A high cohesion degree usually helps to improve the clustering effect. A low cohesion degree means that there are large differences among the data points within the cluster, and the clustering effect is poor. The large distance between data points may indicate that the clustering is not accurate enough.

[0035] Furthermore, a high separation degree (a higher Separation value) means that there are large differences between different clusters. The large distance between the cluster centers indicates that the clustering result is more clearly distinguishable and has a better effect. This means that the boundaries between different clusters are clear, and the clustering model can effectively distinguish different samples. A low separation degree (a lower Separation value) means that the differences between different clusters are small, and the clustering effect is poor. The close distance between the cluster centers may lead to a fuzzy clustering result and poor classification effect.

[0036] Finally, a high evaluation result of the clustering effect means that the clustering result has both strong cohesion (tight within the cluster) and strong separation (clear between clusters), and the clustering effect is good. A low evaluation result of the clustering effect means that the clustering result has poor cohesion (large differences among the data points within the cluster) and poor separation (unclear distinction between clusters), and the clustering effect is poor.

[0037] S3. According to the optimal clustering model, construct a load behavior map, map the current load data to the load behavior map, and adjust the initial threshold interval in combination with the historical load data fluctuation range to obtain the corrected threshold interval.

[0038] In this embodiment, according to the optimal clustering model, construct a load behavior map as follows: Evaluate the clustering effects of each clustering model to obtain the best clustering model; Obtain the historical load data of the area to be measured, and perform sliding window segmentation on this data. Each window contains a time series sample with a preset length; Input the dimensional tensor into the clustering algorithm for training, and output the load behavior category to which each time window belongs. The load behavior categories include but are not limited to peak load mode, valley load mode, periodic load mode, and burst load mode; Cluster all sample tensors through the clustering algorithm to obtain several load behavior categories. Each category constitutes a typical load map, and each map contains its cluster center feature vector and the corresponding load fluctuation range.

[0039] In this embodiment, the current load data is mapped to a load behavior atlas, and the initial threshold interval is adjusted in combination with the historical load data fluctuation range, as follows: Extract the features of the real-time load data within the sliding window at the current moment to form a tensor with the same dimension as the load behavior atlas for similarity matching; Based on the cosine similarity method, obtain the similarity between the current feature vector and the central vectors of all load behavior atlases, and select the load behavior category atlas with the highest similarity as the mapping atlas for the current window; Extract the maximum deviation values within the historical load fluctuation range of this category from the mapping atlas, which are respectively used to represent the upper offset correction amount and the lower offset correction amount as correction factors; Control the initial threshold interval through the correction factors and in combination with the standard deviation of the sliding window load data to obtain the corrected threshold interval.

[0040] The corrected dynamic threshold model is as follows:

[0041]

[0042] In the formula: is the load upper limit, is the predicted load value at the next moment, is the load upper limit, is a predetermined constant coefficient, which is adjusted according to the tolerance of load fluctuation, is the standard deviation of the load data within the sliding window, is the empirical positive direction offset correction amount recorded in the mapped atlas, is the empirical negative direction offset correction amount recorded in the mapped atlas.

[0043] S4. Extract the distribution feature vector and classification accuracy parameter of the current load data as the first data, train the optimal classification model based on the first data, and input the corrected threshold interval into the optimal classification model to identify the current load status label. The first data includes distribution features and classification accuracy.

[0044] In this embodiment, extract the distribution feature vector and classification accuracy parameter of the current load data as the first data, and train the optimal classification model based on the first data, as follows: Obtain the first data of the load data, and the first data includes distribution features and classification accuracy; Obtain the load data of the area to be measured, obtain the distribution features of the load data, and analyze the distribution pattern of the load data through the histogram method; Screen the load data features based on the distribution characteristics of the load data, and construct a subset of load data features; Use the subset of load data features as time series samples to input into a classification model for classification, and evaluate the classification accuracy of the classification model based on cross-validation, and select the best classification model according to the classification accuracy.

[0045] In this embodiment, obtain the load data of the area to be measured, and obtain the first data of the load data. The first data includes distribution characteristics and expected classification accuracy, as follows: Obtain the load data of the area to be measured, obtain the distribution characteristics of the load data, and analyze the distribution pattern of the load data through the histogram method. The distribution characteristics include skewness and kurtosis, and the distribution pattern includes normal distribution and skewed distribution; Screen the load data features based on the distribution characteristics of the load data, and construct a subset of load data features. The load data features include time features and load characteristics; Use the subset of load data features as time series samples to input into a classification model for classification, and evaluate the classification accuracy of the classification model based on cross-validation, and select the best classification model according to the classification accuracy. The classification models include decision tree, support vector machine, K-nearest neighbor, and neural network.

[0046] In this embodiment, screening the load data features based on the distribution characteristics of the load data is as follows: Obtain the distribution pattern of the load data. The distribution pattern includes normal distribution, skewed distribution, and multimodal distribution; According to the distribution type, extract the feature indicators related to the data distribution pattern. The feature indicators include mean, variance, skewness, kurtosis, maximum value, and minimum value; Based on the feature importance ranking method of correlation analysis, evaluate the contribution degree of each feature indicator to the classification performance of the load data; According to the evaluation results, screen out the feature indicators that have the best effect on improving the classification accuracy under the current load data distribution characteristics, and construct a subset of load data features.

[0047] It should be noted that the adaptive ability of the system and the subsequent tuning efficiency. The higher the score, the stronger the adaptability of the current feature selection and clustering strategy, and the faster the system can enter the stable state. If the score is low, the system needs to repeatedly adjust the clustering parameters, feature set or reselect the model, which increases the operation and maintenance cost and the model iteration cycle.

[0048] Therefore, in this embodiment, by introducing a performance score evaluation mechanism to optimize the candidate clustering model, it can effectively improve the recognition accuracy of the system for the load state and the control response speed, thereby enhancing the intelligence, reliability and energy-saving effect of the distribution system operation, and having significant practical value and engineering promotion significance.

[0049] In this embodiment, the corrected threshold interval is input into the optimal classification model to identify the current load status label, specifically as follows: Cluster each data point through the best clustering model and output the initial load status label; Input the load feature data of the current window into the trained classification model to obtain the predicted load status label. The classification model includes a decision tree model, a support vector machine model, a K-nearest neighbor model, or a neural network model; Perform label fusion on the output results of the clustering model and the classification model. The fusion methods include weighted voting method, confidence-weighted average method, or label consistency test method; Take the fused load status label as the final recognition result; According to the finally recognized load status label, perform a control operation on the load switch of the target distribution box.

[0050] The load status label usually includes: Normal load: The load fluctuation is within the preset normal range, and no intervention measures are required; Overload state: The load fluctuation exceeds the preset safety threshold, and the system needs to automatically take measures, such as switching the load switch to avoid system damage; Light load state: The load fluctuation is small, and the system can perform energy-saving regulation according to needs to optimize the load usage efficiency; The control operation strategy for the load switch of the target distribution box is as follows: Normal load: Maintain the current load status and keep the power distribution system running normally Overload state: When the load status is overload, the system immediately controls the target load switch to cut off or adjust the power supply to protect electrical equipment from damage; Light load state: When the load status is light load, the system can adjust the switch of the load circuit according to demand to improve the system operation efficiency.

[0051] The above formulas are all dimensionless and take their numerical values for calculation. The formula is obtained by collecting a large amount of data for software simulation to get a formula closest to the actual situation. The preset parameters in the formula are set by those skilled in the art according to the actual situation.

[0052] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product.

[0053] Those of ordinary skill in the art will appreciate that the modules and algorithm steps of each example described in connection with the embodiments disclosed herein can be implemented in electronic hardware, or in a combination of computer software and electronic hardware. Whether these functions are executed in hardware or software depends on the specific application and design constraints of the technical solution. Skilled artisans may use different methods for each specific application to implement the described functions, but such implementation should not be considered to exceed the scope of this application.

[0054] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically alone, or two or more modules can be integrated into one module.

[0055] As described above, the above are only specific embodiments of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art within the technical scope disclosed in this application can easily think of changes or substitutions, which should all be covered within the protection scope of this application. Therefore, the protection scope of this application should be subject to the protection scope of the claims.

[0056] Finally: The above are only the preferred embodiments of the present invention and are not used to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. A control method for a load switch of a distribution box, characterized in that, It includes the following steps: Obtain the real-time load data of the area to be measured, and build a dynamic switch threshold control model based on LSTM to generate an initial threshold interval; According to the optimal clustering model, construct a load behavior map, map the current load data to the load behavior map, and adjust the initial threshold interval in combination with the historical load data fluctuation interval to obtain a corrected threshold interval; Extract the distribution feature vector and classification accuracy parameter of the current load data as the first data, and train the optimal classification model based on the first data; Input the corrected threshold interval into the optimal classification model to identify the current load status label.

2. The load switch control method of the distribution box according to claim 1, wherein, The step of obtaining the real-time load data of the area to be measured and building a dynamic switch threshold control model based on LSTM to generate an initial threshold interval is specifically as follows: Monitor each phase line and loop of the distribution box in the area to be measured in real time, synchronously collect the electrical parameter data of each loop at a fixed period to form multi-channel parallel load data; Preprocess the real-time load data through an edge computing node, and perform time synchronization on each data stream based on the NTP protocol; Organize the preprocessed load data in chronological order, and use the sliding window technique to perform time series recombination. The load data of every N consecutive time points constitutes a single time series sample, and each sample constitutes a dimensional tensor; Input the time series sample into the LSTM model for training to obtain a trained dynamic switch threshold model, and predict the load fluctuation trend based on the LSTM model; Construct a dynamic threshold range according to the predicted value output by the LSTM model combined with the dynamic switch threshold model.

3. The load switch control method of the distribution box according to claim 2, characterized in that, The step of constructing a dynamic threshold range according to the predicted value output by the LSTM model combined with the dynamic switch threshold model is specifically as follows: Based on the load data within the window, output the standard deviation of the load data within the window; Construct a dynamic threshold model according to the predicted load trend and the standard deviation of the load data within the sliding window to obtain a dynamic threshold range.

4. The load switch control method for the distribution box according to claim 3, characterized in that, The optimal clustering model is specifically as follows: Preprocess the load data, and reduce the dimension of the load feature data based on the principal component analysis method to obtain sample data; Based on the locality-sensitive hashing algorithm, map similar sample data to the same bucket through hash coding to generate hash values; Cluster the hash values, and based on the cosine similarity method, obtain the local neighborhood of each data point; Extract local features according to the data in the local neighborhood; Evaluate the clustering effect of the clustering model according to the local features, and select the best clustering model.

5. The load switch control method of the distribution box according to claim 4, characterized in that, The step of constructing a load behavior map according to the optimal clustering model is specifically as follows: Obtain the historical load data of the area to be measured, and perform sliding window segmentation. Each window contains a time series sample of a preset length; Construct tensors from the time series samples, and input the tensors into the clustering model for training to output the load behavior category to which each time window belongs; Cluster the sample tensors through the trained clustering model. According to the clustering results, group the time window data belonging to the same category into the same group to obtain several load behavior categories, and each category constitutes a load behavior map.

6. The method for controlling the load switch of the distribution box according to claim 5, characterized in that, Mapping the current load data to the load behavior atlas and adjusting the initial threshold interval in combination with the historical load data fluctuation range are as follows: Extract the features of the real-time load data within the sliding window at the current moment to form a tensor with the same dimension as the load behavior atlas; Based on the cosine similarity method, obtain the similarity between the current feature vector and the central vectors of all load behavior atlases, and select the load behavior category atlas with the highest similarity as the mapping atlas for the current window; Extract the maximum deviation values within the historical load fluctuation range of this category from the mapping atlas, and use them to represent the upper offset correction amount and the lower offset correction amount respectively as correction factors; Control the initial threshold interval through the correction factors in combination with the standard deviation of the sliding window load data to obtain the corrected threshold interval.

7. The load switch control method of the distribution box according to claim 6, characterized in that Extracting the distribution feature vector and classification accuracy parameter of the current load data as the first data, and training the optimal classification model based on the first data are as follows: Obtain the first data of the load data, and the first data includes distribution features and classification accuracy; Obtain the load data of the area to be measured, obtain the distribution features of the load data, and analyze the distribution pattern of the load data through the histogram method; Screen the load data features based on the distribution features of the load data and construct a load data feature subset; Take the load data feature subset as a time series sample and input it into the classification model for classification, and evaluate the classification accuracy of the classification model based on cross-validation, and select the best classification model according to the classification accuracy.

8. The load switch control method for the distribution box according to claim 7, wherein The screening of the load data features based on the distribution features of the load data is as follows: Obtain the distribution pattern of the load data and extract the feature indicators related to the data distribution pattern; Based on the feature importance ranking method of correlation analysis, evaluate the contribution degree of each feature indicator to the load data classification performance; According to the evaluation results, screen out the feature indicators that have the best effect on improving the classification accuracy under the current load data distribution features, and construct a load data feature subset.

9. The load switch control method of the distribution box according to claim 8, characterized in that, Inputting the corrected threshold interval into the optimal classification model to identify the current load status label is as follows: Cluster each data point through the best clustering model and output the initial load status label; Input the load feature data of the current window into the trained classification model to obtain the predicted load status label; Fuse the output results of the clustering model and the classification model; Take the fused load status label as the final recognition result and control the load switch of the target distribution box.

Images