A distribution network adaptive reclosing optimization method and system

Abstract

The present invention relates to the field of reclosing technology, and specifically to a distribution network adaptive reclosing optimization method and system, the method comprising the following steps: S1: collecting historical fault data of the distribution network, and preprocessing the historical fault data to generate three-dimensional time series data; S2: extracting key features based on the three-dimensional time series data using a multi-scale convolutional neural network, and processing the key features to obtain input features; S3: establishing a Transformer network model, classifying the input features to obtain samples, analyzing the relationship information of the samples to obtain an optimization model and classify the fault areas, generating classification labels and predicting the fault duration; S4: for non-permanent faults: using the optimization model to dynamically adjust the reclosing delay time and action sequence. The present invention optimizes the time and sequence of reclosing actions by analyzing fault data through deep learning.

Description

Technical Field

[0001] The present invention relates to the field of reclosing technology, and more particularly to a distribution network adaptive reclosing optimization method and system. Background Art

[0002] In distribution networks, reclosing devices are critical equipment for ensuring power continuity and system stability. Their core function is to quickly cut off the current and attempt to restore power after a fault occurs. Traditional reclosing technology relies primarily on mechanical relays and fixed sequential logic, using preset reclosing times and delay times (such as 0.5 seconds or 15 seconds) to handle transient faults (such as lightning strikes and tree branch shorts).

[0003] However, with the expansion of distribution network scale, the increase in the proportion of new energy access and the diversification of load types, traditional technologies have a significant defect of misjudging fault types, that is, they are unable to distinguish between permanent faults (such as equipment breakdown) and non-permanent faults (such as instantaneous arcs), resulting in repeated reclosing under permanent faults, increasing the risk of equipment damage. Although the existing technology can distinguish faults, it lacks a process for optimizing the time and timing of reclosing actions according to the fault type. Summary of the Invention

[0004] The purpose of the present invention is to overcome the shortcomings of the prior art in misjudging the type of reclosing fault, and to propose a distribution network adaptive reclosing optimization method and system, which can accurately judge the fault type and optimize the time and sequence of reclosing.

[0005] To achieve the above object, a distribution network adaptive reclosing optimization method of the present invention comprises the following steps:

[0006] S1: Collect historical fault data of the distribution network and pre-process the historical fault data to generate three-dimensional time series data. The historical fault data includes waveform data of voltage, current, and power factor collected from multiple sensors of the distribution network at the time of historical faults;

[0007] S2: Extract key features using a multi-scale convolutional neural network based on the three-dimensional time series data, and process the key features to obtain input features, where the key features include fluctuation frequency and amplitude change;

[0008] S3: Establish a Transformer network model, classify the input features to obtain samples, analyze the relationship information of the samples to obtain an optimized model, and classify the faults into permanent faults and non-permanent faults, generate classification labels, and predict the fault duration;

[0009] S4: For non-permanent faults: dynamically adjust the reclosing delay time using the optimization model to optimize the reclosing action time; divide the timing window, and adaptively adjust the reclosing action timing based on the load intensity and the optimization model to optimize the reclosing action timing.

[0010] In this technical solution, historical fault data is collected from multiple sensors (current transformers, voltage transformers, etc.) in the distribution network. The data is cleaned and abnormal data is processed to improve the accuracy of subsequent judgments. Different types of faults are manually identified and labeled (permanent and non-permanent) in preparation for deep learning. After processing, the data is flattened and spliced into three-dimensional time series data. This three-dimensional time series data is: data with three dimensions (number of samples, time steps, and number of features) arranged in chronological order, which reflects the changes in historical faults in the distribution network over time. The three-dimensional time series data can be processed by a deep learning model. A convolution operation is performed on the three-dimensional time series data to extract features to obtain input and output features. A Transformer network model is established, and the input features are classified to obtain samples. The relationship information between the samples is analyzed to obtain an optimization model. The optimization model is used to classify the fault type. If the fault is non-permanent, a reclosing operation is triggered, an optimal delay time is set, and the reclosing action timing is optimized to improve the reclosing success rate. If the fault is permanent, the reclosing restart is avoided to reduce unnecessary system disturbances.

[0011] As a preferred solution, step S1 includes:

[0012] S11: Collecting historical record data of each reclosing operation; the waveform data includes current waveform, voltage waveform, and duration;

[0013] S12: Cleaning the waveform data and the historical record data of each reclosing operation, and processing abnormal data;

[0014] S13: Label different types of faults at the time of power outage;

[0015] S14: The processed data is flattened and spliced into the three-dimensional time series data.

[0016] As a preferred solution, step S2 includes:

[0017] S21: using a multi-scale convolutional neural network to perform a convolution operation on the waveform data and extract key features; the key features are the fluctuation frequency and amplitude changes in the reclosing waveform;

[0018] S22: Add a time series convolution layer to the multi-scale convolutional neural network to obtain a time series model. Input key features into the time series model to obtain multi-time step sequence features to describe the dynamic change characteristics when a reclosing fault occurs.

[0019] S23: applying convolution kernels of different scales to the sequence features of the multiple time steps to obtain features in different time windows to form multi-scale features;

[0020] S24: adding position encoding to the multi-scale features to provide position information and obtain input features.

[0021] As a preferred solution, in step S23, the calculation formula of the multi-scale feature is:

[0022]

[0023] Where: is a multi-scale feature, is the feature splicing operation, 、 and For input Perform convolutions of different scales.

[0024] As a preferred solution, step S3 includes:

[0025] S31: Establish a Transformer network model, classify the input features based on a multi-head self-attention mechanism to obtain different samples, calculate the relationship scores between different samples, and determine the similarity between samples. The relationship scores between samples of the same category are higher, while the relationship scores between samples of different categories are lower. The self-attention mechanism captures the relationship information between samples and can effectively capture the correlation between different time steps. The Transformer network model learns fault features by encoding the correlation between different time points, thereby improving the accuracy of fault classification and the robustness of the model.

[0026] S32: Based on the cross-attention mechanism, the dependency relationship between different time steps is optimized according to the following formula to obtain an optimized model:

[0027]

[0028] Where: represents the input data at different time steps, represents the query matrix, Represents the learnable weight matrix , represents the bond matrix, Represents the learnable weight matrix , represents the value matrix, Represents the learnable weight matrix ;

[0029] S33: Use the multi-layer perceptron of the optimized model to distinguish permanent from non-permanent faults, generate classification labels and predict fault duration.

[0030] As a preferred solution, in step S4, the process of optimizing the reclosing action time is as follows: based on the fault classification result, the reclosing delay time is dynamically adjusted using the characteristics and prediction results of the optimization model; if it is determined to be a non-permanent fault, the reclosing delay time is adjusted to ensure that the reclosing operation is performed after the arc is completely dissociated, thereby improving the success rate. The calculation formula of the adaptive reclosing delay time is:

[0031]

[0032] Where: is the adaptive reclosing delay time, is the basic delay time, Based on the fault type The control function of is the adjustment factor.

[0033] As a preferred solution, in step S4, the process of optimizing the reclosing action sequence is as follows: defining multiple time windows to ensure that the reclosing action within each window has the best activation and delay sequence, and setting the duration and delay time of multiple reclosing attempts based on the load intensity and the prediction parameters of the optimization model. The calculation formula for optimizing the reclosing action sequence is:

[0034]

[0035] Where: is the action time of the i-th reclosing circuit breaker, is the basic delay of the i-th time window, is the delay sequence calculated based on different windows, is the adjustment factor.

[0036] As a preferred solution, it also includes step S5: adaptively optimizing the reclosing strategy by updating the parameters of the optimization model through real-time data stream. This step can realize adaptive tuning by updating the deep learning data online.

[0037] As a preferred solution, step S5 includes: incrementally updating the parameters of the optimization model according to the new fault data, adaptively optimizing the reclosing delay time and action timing strategy, and the optimization model adopts an adaptive control strategy, which can update the data in real time to adapt to the power load fluctuations and environmental changes in the distribution network.

[0038] To achieve the above-mentioned object, the present invention provides a distribution network adaptive reclosing optimization system, which is used to implement the above-mentioned distribution network adaptive reclosing optimization method, comprising:

[0039] Data acquisition module: used to collect historical fault data of the distribution network, and pre-process the historical fault data to generate three-dimensional time series data;

[0040] Feature extraction module: used to extract key features from the three-dimensional time series data and process the key features to obtain input features;

[0041] Feature classification module: used to establish a network model, classify the input features to obtain samples, analyze the relationship information of the samples to obtain an optimized model and classify faults;

[0042] Action parameter optimization module: used to dynamically optimize the reclosing action time according to the fault classification result and using the optimization model, and adaptively adjust the reclosing action timing based on the load intensity and the optimization model.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] 1. This invention improves the accuracy of fault classification by performing convolution operations on fault data to extract features. It uses the self-attention mechanism to optimize the model, accurately classifies faults into permanent and non-permanent, and then optimizes the reclosing strategy based on the judgment results, effectively improving the flexibility and accuracy of reclosing and avoiding equipment damage caused by repeated reclosing under permanent faults.

[0045] 2. By optimizing the features and prediction results provided by the model, the reclosing delay time is dynamically adjusted according to the fault classification results, achieving adaptive reclosing action time optimization, improving the reclosing success rate under non-permanent faults, and effectively reducing the impact of faults on power supply.

[0046] 3. Real-time adjustment of reclosing strategies under different loads and environments enhances the stability and security of the power grid in complex scenarios. By optimizing the reclosing action sequence, it accelerates fault isolation and recovery of non-fault areas, significantly reduces power outage time for users, and improves the overall power grid service level. At the same time, it can avoid unnecessary current shocks, reduce the possibility of secondary faults, and ensure system safety.

[0047] 4. Incrementally update the optimization model parameters based on new fault data, and adaptively optimize the reclosing delay time and action sequence to adapt to power load fluctuations and environmental changes in the distribution network. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 It is a flow chart of the distribution network adaptive reclosing optimization method of the present invention;

[0049] Figure 2 It is a flow chart of data processing of the present invention.

[0050] Figure 3 Schematic diagram of the distribution network adaptive reclosing optimization system of the present invention. DETAILED DESCRIPTION

[0051] The drawings are for illustrative purposes only and should not be construed as limiting this patent. To better illustrate the embodiments, some components in the drawings may be omitted, enlarged, or reduced in size, and do not represent actual product dimensions. Those skilled in the art will understand that some well-known structures and their descriptions may be omitted from the drawings. The positional relationships depicted in the drawings are for illustrative purposes only and should not be construed as limiting this patent.

[0052] The same or similar numbers in the drawings of the embodiments of the present invention correspond to the same or similar parts; in the description of the present invention, it should be understood that if there are terms such as "upper", "lower", "left", "right", "long", "short", etc. indicating the orientation or position relationship, they are based on the orientation or position relationship shown in the drawings. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operate in a specific orientation. Therefore, the terms describing the position relationship in the drawings are only used for illustrative purposes and cannot be understood as limiting this patent. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to specific circumstances.

[0053] The technical solution of the present invention is further described in detail below through specific embodiments and in conjunction with the accompanying drawings:

[0054] Example 1:

[0055] like Figure 1 、 2 As shown, this embodiment provides a distribution network adaptive reclosing optimization method, including the following steps:

[0056] S1: Collect historical fault data of the distribution network and pre-process the historical fault data to generate three-dimensional time series data. The historical fault data includes waveform data of voltage, current, and power factor collected from multiple sensors of the distribution network at the time of historical faults;

[0057] S2: Extract key features using a multi-scale convolutional neural network based on the three-dimensional time series data, and process the key features to obtain input features, where the key features include fluctuation frequency and amplitude change;

[0058] S3: Establish a Transformer network model, classify the input features to obtain samples, analyze the relationship information of the samples to obtain an optimized model, and classify the faults into permanent faults and non-permanent faults, generate classification labels, and predict the fault duration;

[0059] S4: For non-permanent faults: dynamically adjust the reclosing delay time using the optimization model to optimize the reclosing action time; divide the timing window, and adaptively adjust the reclosing action timing based on the load intensity and the optimization model to optimize the reclosing action timing.

[0060] In this embodiment, after a fault occurs, the optimization model can quickly determine the fault type. If it is a non-permanent fault, the reclosing operation is triggered and the optimal delay time and action sequence are set to improve the reclosing success rate. If it is a permanent fault, the reclosing restart is avoided to reduce unnecessary system disturbances.

[0061] Specifically, step S1 includes:

[0062] S11: Collecting historical record data of each reclosing operation; the waveform data includes current waveform, voltage waveform, and duration;

[0063] S12: Cleaning the waveform data and the historical record data of each reclosing operation, and processing abnormal data;

[0064] S13: Label different types of faults at the time of power outage;

[0065] S14: The processed data is flattened and spliced into the three-dimensional time series data.

[0066] In this embodiment, the three-dimensional time series data is data with three dimensions (number of samples, time steps, number of features) arranged in chronological order.

[0067] Specifically, step S2 includes:

[0068] S21: Using a multi-scale convolutional neural network to perform a convolution operation on the waveform data and extract key features;

[0069] S22: Add a temporal convolution layer to the multi-scale convolutional neural network to obtain a time series model, and input key features into the time series model to obtain multi-time step sequence features;

[0070] S23: applying convolution kernels of different scales to the sequence features of the multiple time steps to obtain features in different time windows to form multi-scale features;

[0071] S24: Add position encoding to the multi-scale features to obtain input features.

[0072] In this embodiment, the input formula of the multi-scale convolutional neural network model is:

[0073]

[0074] Where: For input data, is a real matrix, is the number of time steps, is the spatial resolution, is the number of channels.

[0075] In this embodiment, the calculation formula of the multi-scale feature is:

[0076]

[0077] Where: is a multi-scale feature, is the feature splicing operation, 、 and For input Perform convolutions of different scales.

[0078] In this embodiment, after adding the position code, the position where the fault occurs can be marked in the fault feature to form a new optimization model input, wherein the calculation formula of the input is:

[0079]

[0080] Where: Represents multi-scale features, represents the position code, Indicates the input after adding location information.

[0081] In this embodiment, the intelligent reclosing device can be connected to the distribution automation system (DAS) and work in conjunction with the SCADA system to achieve centralized monitoring and remote control. It can automatically select the reclosing method and time based on the fault location, equipment status and load properties, effectively improving the flexibility and accuracy of the reclosing device.

[0082] Specifically, step S3 includes:

[0083] S31: Establish a Transformer network model, classify the input features based on a multi-head self-attention mechanism to obtain different samples, calculate the relationship scores between different samples, and determine the similarity between the samples. The relationship scores between samples of the same category are higher, while the relationship scores between samples of different categories are lower. The multi-head self-attention mechanism is used to capture the relationship information between samples, thereby improving the classification accuracy and the robustness of the model.

[0084] S32: Based on the cross-attention mechanism, the dependency relationship between different time steps is optimized according to the following formula to obtain an optimized model:

[0085]

[0086] Where: represents the input data at different time steps, represents the query matrix, Represents the learnable weight matrix , represents the bond matrix, Represents the learnable weight matrix , represents the value matrix, Represents the learnable weight matrix ;

[0087] S33: Use the multi-layer perceptron of the optimized model to distinguish permanent from non-permanent faults, generate classification labels and predict fault duration.

[0088] In this embodiment, in step S34, the calculation formula for the query, key and value in the cross attention is:

[0089]

[0090] Where: represents the input data at different time steps, represents the query matrix, Represents the learnable weight matrix , represents the bond matrix, Represents the learnable weight matrix , represents the value matrix, Represents the learnable weight matrix .

[0091] In this embodiment, the calculation formula of self-attention in the encoder is:

[0092] Where: represents the output of the attention calculation, is the similarity score of the two matrices, is a nonlinear activation function, is the dimension of the Key vector.

[0093] In this embodiment, the output passes through the fully connected layer and finally passes through the classifier to obtain the output:

[0094]

[0095] Where: For output, It is a fully connected layer, which is mapped to the output space of the target task through the MLP layer for classification.

[0096] Specifically, in step S4, the process of optimizing the reclosing action time is: according to the fault classification result, the characteristics and prediction results of the optimization model are used to dynamically adjust the reclosing delay time to ensure that the reclosing operation is performed after the arc is completely dissociated, thereby improving the success rate.

[0097] In this embodiment, the calculation formula of the adaptive reclosing delay time is:

[0098]

[0099] Where: is the adaptive reclosing delay time, is the basic delay time, Based on the fault type The control function of is the adjustment factor.

[0100] Specifically, in step S4, the process of optimizing the reclosing action sequence is as follows: defining multiple time windows to ensure that the reclosing action within each window has the best activation and delay sequence, and setting the duration and delay time of multiple reclosing attempts based on the load intensity and the prediction parameters of the optimization model. The calculation formula for optimizing the reclosing action sequence is:

[0101]

[0102] Where: is the action time of the i-th reclosing circuit breaker, is the basic delay of the i-th time window, is the delay sequence calculated based on different windows, is the adjustment factor.

[0103] Example 2:

[0104] This embodiment is similar to embodiment 1, except that, in this embodiment, Figure 1 As shown, the method further includes step S5: incrementally updating the parameters of the optimization model according to the new fault data, and adaptively optimizing the reclosing delay time and action timing strategy.

[0105] Specifically, the step S5 includes: incrementally updating the parameters of the optimization model according to the new fault data, and adaptively optimizing the reclosing delay time and action timing strategy.

[0106] In this embodiment, through online updating and adaptive tuning, the reclosing delay time and action timing strategy can be adjusted and optimized in real time to adapt to power load fluctuations and environmental changes in the distribution network.

[0107] Example 3:

[0108] like Figure 3 As shown, this embodiment provides a distribution network adaptive reclosing optimization system for implementing the above-mentioned distribution network adaptive reclosing optimization method, including:

[0109] Data acquisition module: used to collect historical fault data of the distribution network, and pre-process the historical fault data to generate three-dimensional time series data;

[0110] Feature extraction module: used to extract key features from the three-dimensional time series data and process the key features to obtain input features;

[0111] Feature classification module: used to establish a network model, classify the input features to obtain samples, analyze the relationship information of the samples to obtain an optimized model and classify faults;

[0112] Action parameter optimization module: used to dynamically optimize the reclosing action time according to the fault classification result and using the optimization model, and adaptively adjust the reclosing action timing based on the load intensity and the optimization model.

[0113] In this embodiment, during the operation of the distribution network adaptive reclosing optimization system, the optimized model parameters are updated in real time and input into the power system distribution network model containing the reclosing device to achieve adaptive optimization.

[0114] Although the embodiments of the present invention have been shown and described above, it will be understood that the above embodiments are exemplary and are not to be construed as limitations on the present invention. A person skilled in the art may change, modify, replace and modify the above embodiments within the scope of the present invention.

[0115] Obviously, the above embodiments of the present invention are merely examples for the purpose of clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. Those skilled in the art will appreciate that other variations or modifications can be made based on the above description. It is not necessary and impossible to enumerate all embodiments here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the claims of the present invention.

Claims

1. A distribution network adaptive reclosing optimization method, characterized in that: The following steps are involved: S1: Collect historical fault data of the distribution network and pre-process the historical fault data to generate three-dimensional time series data. The historical fault data includes waveform data of voltage, current, and power factor collected from multiple sensors of the distribution network at the time of historical faults; S2: Extract key features using a multi-scale convolutional neural network based on the three-dimensional time series data, and process the key features to obtain input features, where the key features include fluctuation frequency and amplitude change; S3: Establish a Transformer network model, classify the input features to obtain samples, analyze the relationship information of the samples to obtain an optimized model, and classify the faults into permanent faults and non-permanent faults, generate classification labels, and predict the fault duration; S4: For non-permanent faults: dynamically adjust the reclosing delay time using the optimization model to optimize the reclosing action time. The process of optimizing the reclosing action time is as follows: dynamically adjust the reclosing delay time based on the fault classification result using the characteristics and prediction results of the optimization model. The calculation formula for the adaptive reclosing delay time is: Where: is the adaptive reclosing delay time, is the basic delay time, Based on the fault type The control function of The adjustment coefficient is obtained by dividing the timing window and adaptively adjusting the reclosing action timing based on the load intensity and the optimization model to optimize the reclosing action timing. The process of optimizing the reclosing action timing is as follows: defining multiple time windows, and setting the duration and delay time of multiple reclosing attempts according to the load intensity and the prediction parameters of the optimization model. The calculation formula for optimizing the reclosing action timing is: Where: is the action time of the i-th reclosing circuit breaker, is the basic delay of the i-th time window, is the delay sequence calculated based on different windows, is the adjustment factor; If it is a permanent fault: avoid reclosing and restarting to reduce unnecessary system disturbances.

2. A distribution network adaptive reclosing optimization method according to claim 1, characterized in that: Step S1 includes: S11: Collecting historical record data of each reclosing operation; the waveform data includes current waveform, voltage waveform, and duration; S12: Cleaning the waveform data and the historical record data of each reclosing operation, and processing abnormal data; S13: Label different types of faults at the time of power outage; S14: The processed data is flattened and spliced into the three-dimensional time series data.

3. A distribution network adaptive reclosing optimization method according to claim 2, characterized in that: Step S2 includes: S21: Using a multi-scale convolutional neural network to perform a convolution operation on the waveform data and extract key features; S22: Add a temporal convolution layer to the multi-scale convolutional neural network to obtain a time series model, and input key features into the time series model to obtain multi-time step sequence features; S23: applying convolution kernels of different scales to the sequence features of the multiple time steps to obtain features in different time windows to form multi-scale features; S24: Add position encoding to the multi-scale features to obtain input features.

4. A distribution network adaptive reclosing optimization method according to claim 3, characterized in that: In step S23, the calculation formula of the multi-scale feature is: Where: is a multi-scale feature, is the feature splicing operation, 、 and For input Perform convolutions of different scales.

5. A distribution network adaptive reclosing optimization method according to claim 1, characterized in that: Step S3 includes: S31: Establish a Transformer network model, classify the input features based on the multi-head self-attention mechanism to obtain different samples, calculate the relationship scores between different samples, and determine the similarity between the samples; S32: Based on the cross-attention mechanism, the dependency relationship between different time steps is optimized according to the following formula to obtain an optimized model: Where: represents the input data at different time steps, represents the query matrix, Represents the learnable weight matrix , represents the bond matrix, Represents the learnable weight matrix , represents the value matrix, Represents the learnable weight matrix ; S33: Use the multi-layer perceptron of the optimized model to distinguish permanent from non-permanent faults, generate classification labels and predict fault duration.

6. A distribution network adaptive reclosing optimization method according to any one of claims 1 to 5, characterized in that: The method further includes step S5: updating the parameters of the optimization model through real-time data stream to adaptively optimize the reclosing strategy.

7. A distribution network adaptive reclosing optimization method according to claim 6, characterized in that: The step S5 includes: incrementally updating the parameters of the optimization model according to the new fault data, and adaptively optimizing the reclosing delay time and action timing strategy.

8. A distribution network adaptive reclosing optimization system, characterized in that: A method for optimizing the adaptive reclosing of a distribution network according to any one of claims 1 to 7, comprising: Data acquisition module: used to collect historical fault data of the distribution network, and pre-process the historical fault data to generate three-dimensional time series data; Feature extraction module: used to extract key features from the three-dimensional time series data and process the key features to obtain input features; Feature classification module: used to establish a network model, classify the input features to obtain samples, analyze the relationship information of the samples to obtain an optimized model and classify faults; Action parameter optimization module: used to dynamically optimize the reclosing action time according to the fault classification result and using the optimization model, and adaptively adjust the reclosing action timing based on the load intensity and the optimization model.

Images