Power distribution network power failure fault prediction method based on digital twinborn model

Abstract

The invention discloses a power distribution network power failure prediction method based on a digital twinborn model, and the method comprises the steps: collecting parameter data related to a power failure in a power distribution network, and the parameter data comprises an electrical parameter, an equipment state parameter, meteorological data, power grid operation information and a historical power failure record; constructing a power distribution network digital twinborn model according to the preprocessed parameter data and the structure information of the power distribution network; an XGBoost algorithm is adopted to train the multi-source time sequence data of the twin digital model of the power distribution network, and a power failure prediction model is generated; and inputting the real-time operation parameter data of the power distribution network into the power failure prediction model for prediction, and outputting a power failure prediction result of the power distribution network.

Description

Technical Field

[0001] The present invention relates to the technical field of distribution network operation management, and more specifically, to a distribution network power outage fault prediction method based on a digital twin model. Background Art

[0002] In the operation of distribution networks, power outages can have a serious impact on social production and life. Accurate power outage fault analysis helps power companies take measures in advance to reduce power outage losses and improve power supply reliability. The causes of power outages are complex and diverse, including natural disasters, equipment failures, and human operational errors. For example, extreme weather may cause damage to transmission lines, causing large-scale power outages; aging equipment may cause local failures, affecting power supply in specific areas. Therefore, an effective method is needed to comprehensively analyze various factors and achieve accurate prediction of power outages.

[0003] At present, the power outage fault analysis methods mainly include those based on statistical models, machine learning models, etc. However, these methods have certain limitations. The statistical model-based methods have strict assumptions on data distribution and are difficult to adapt to the complex and changeable distribution network environment. Although machine learning models such as neural networks have improved the prediction accuracy to a certain extent, they are not effective in feature extraction when processing time series data, and lack effective modeling of long-term data dependencies. At the same time, the existing methods are also lacking in real-time performance, making it difficult to meet the distribution network's needs for rapid response to power outages. Summary of the invention

[0004] In view of the deficiencies in the prior art, the present invention provides a distribution network power outage fault prediction method based on a digital twin model.

[0005] According to one aspect of the present invention, a method for predicting power outage faults in a distribution network based on a digital twin model is provided, comprising:

[0006] Collect parameter data related to power outages in the distribution network, including electrical parameters, equipment status parameters, meteorological data, grid operation information, and historical power outage records;

[0007] Construct a digital twin model of the distribution network based on the preprocessed parameter data and the structural information of the distribution network;

[0008] The XGBoost algorithm is used to train the multi-source time series data of the distribution network twin digital model to generate a power outage fault prediction model;

[0009] The real-time operating parameter data of the distribution network is input into the power outage fault prediction model for prediction, and the power outage fault prediction results of the distribution network are output.

[0010] Optionally, the preprocessing operation of the parameter data includes:

[0011] The Z-score method was used to remove outliers in parameter data;

[0012] A low-pass filter is used to remove noise interference in the parameter data to obtain preprocessed parameter data.

[0013] Optionally, it also includes:

[0014] Using an incremental update algorithm, the digital twin model of the distribution network is dynamically updated based on the continuously collected real-time parameter data of the distribution network.

[0015] Optionally, the XGBoost algorithm is used to train the multi-source time series data of the distribution network twin digital model to generate a power outage fault prediction model, including:

[0016] The multi-source time series data of the distribution network twin digital model is input into the convolutional neural network for feature extraction to obtain the local features of the multi-source time series data;

[0017] The local features are input into the long short-term memory network for global feature extraction to obtain the global features of multi-source time series data;

[0018] The global features are input into the attention mechanism layer for attention weighting, and the weighted feature data of multi-source time series are output;

[0019] The weighted feature data is trained and predicted using the XGBoost algorithm to obtain the historical power outage fault prediction results of the distribution network;

[0020] Compare and evaluate the historical power outage fault prediction results with the historical actual power outage fault conditions to determine the model evaluation results;

[0021] According to the model evaluation results, the feedback mechanism is used to optimize and adjust the model to generate a power outage fault prediction model.

[0022] Optionally, the multi-source time series data of the distribution network twin digital model is input into a convolutional neural network for feature extraction to obtain local features of the multi-source time series data, including:

[0023] The initial local features of multi-source time series data are extracted by sliding the convolution kernel on the time series;

[0024] The initial local features are pooled and reconstructed to output local features in a format suitable for long short-term memory network processing.

[0025] Optionally, the convolution operation formula is:

[0026]

[0027] In the formula, x is the feature sequence of multi-source time series data; w is the convolution kernel; i represents the sliding position of the convolution kernel in the time step direction, j and l represent the row and column positions of the convolution kernel on the two-dimensional feature plane respectively;

[0028] The pooling operation formula is:

[0029] (M Pooling (x,ps))[i]=max{x[i+j-1,l],j=1,…,ps,l=1,…,ps}

[0030] Where ps is the pooling window size.

[0031] Optionally, the LSTM input gate i t 、Forget gate t , Update cell status C t and output gate o t ,in,

[0032] Input Gate i t =σ(W xi ·x t +W hi ·h t-1 +b i ), controls the update of cell state, where x t is the input vector at the current moment, h t-1 is the hidden state vector at the previous moment, W xi and W hi is the weight matrix, b i is the bias term, σ represents the Sigmoid activation function;

[0033] Forget Gate t =σ(W xf ·x t +W hf ·h t-1 +b f ), determines the degree of forgetting of the cell state at the previous moment, where W xf and W hf is the weight matrix, b f is bias;

[0034] Update cell status C t =f t ⊙C t-1 +i t ⊙tanh(W xc ·x t +W hc ·h t-1 +b c ), where C t-1is the cell state at the previous moment, tanh represents the hyperbolic tangent activation function;

[0035] Output gate o t =σ(W xo ·x t +W ho ·h t-1 +b o ), controls the mapping between cell state and hidden state, where W xo and W ho is the weight matrix, b o is bias;

[0036] The final hidden state of the LSTM network is h t =o t ⊙tanh(C t ).

[0037] Optionally, the weight matrix A of the attention mechanism layer is calculated as:

[0038] A=softmax(E)

[0039] Where E = tanh(W attention ·H+b attention ), W attention is the learnable parameter matrix, b attention is the bias, and H is the global feature output by the long short-term memory network.

[0040] Optionally, the decision tree ensemble model of the XGBoost algorithm is:

[0041]

[0042] The objective function is:

[0043]

[0044] In the formula, are the true value and predicted value of the nth data respectively; is the e-th decision tree function in the decision tree space F, corresponding to the structure of the e-th tree and the relevant information of the leaf weights; l(·) is the training error function; ω(f e ) is the regular term for controlling complexity, N is the number of input data, and E is the number of decision trees.

[0045] Optionally, historical power outage fault prediction results are compared and evaluated with historical actual power outage fault conditions to determine model evaluation results, including:

[0046] The accuracy and recall of the model are calculated based on the historical power outage prediction results and the actual historical power outage conditions;

[0047] The F1 score is calculated based on the precision and recall to determine the model evaluation result.

[0048] Optionally, the F1 score is calculated as:

[0049]

[0050] in, TP is the true positive, which is the number of times the power outage is correctly predicted and actually occurs; FP is the false positive, which is the number of times the power outage is incorrectly predicted but does not actually occur. FN is the false negative, which is the number of times a power outage actually occurred but was not predicted.

[0051] According to another aspect of the present invention, a distribution network power outage fault prediction device based on a digital twin model is provided, comprising:

[0052] The acquisition module is used to collect parameter data related to power outage faults in the distribution network, where the parameter data includes electrical parameters, equipment status parameters, meteorological data, power grid operation information, and historical power outage fault records;

[0053] A construction module is used to construct a digital twin model of the distribution network based on the preprocessed parameter data and the structural information of the distribution network;

[0054] The training module is used to train the multi-source time series data of the distribution network twin digital model using the XGBoost algorithm to generate a power outage fault prediction model;

[0055] The prediction module is used to input the real-time operating parameter data of the distribution network into the power outage fault prediction model for prediction, and output the power outage fault prediction result of the distribution network.

[0056] According to another aspect of the present invention, a computer-readable storage medium is provided, wherein the storage medium stores a computer program, and the computer program is used to execute the method described in any one of the above aspects of the present invention.

[0057] According to another aspect of the present invention, an electronic device is provided, comprising: a processor; a memory for storing instructions executable by the processor; the processor is configured to read the executable instructions from the memory and execute the instructions to implement the method described in any one of the above aspects of the present invention.

[0058] Therefore, the present invention integrates advanced digital twin technology and deep learning algorithms to construct a highly accurate and real-time dynamically updated digital twin model of the distribution network, which can accurately analyze and predict power outages in advance, and effectively improve the operation management level and power supply reliability of the distribution network system. It can achieve comprehensive, real-time and accurate analysis and prediction of power outages in the distribution network. The close integration of the digital twin model and the actual system makes the monitoring of the distribution network status more accurate and the prediction of power outages more timely and reliable. This helps power companies to formulate response measures in advance, reasonably arrange operation and maintenance resources, quickly restore power supply, effectively reduce the impact of power outages on social production and life, improve the overall operation efficiency and reliability of the distribution network, and ensure the stability and safety of power supply. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] A more complete understanding of exemplary embodiments of the present invention may be obtained by referring to the following drawings:

[0060] Figure 1 It is a flow chart of a method for predicting power outage faults in a distribution network based on a digital twin model provided by an exemplary embodiment of the present invention;

[0061] Figure 2 It is another flow chart of a method for predicting power outage faults in a distribution network based on a digital twin model provided by an exemplary embodiment of the present invention;

[0062] Figure 3 It is a typical structural diagram of a CNN network model provided by an exemplary embodiment of the present invention;

[0063] Figure 4 It is a schematic diagram of the encoder structure of an LSTM model based on a multidimensional time series provided by an exemplary embodiment of the present invention;

[0064] Figure 5 It is a typical structural diagram of an LSTM model unit provided by an exemplary embodiment of the present invention;

[0065] Figure 6 It is a structural schematic diagram of a distribution network power outage fault prediction device based on a digital twin model provided by an exemplary embodiment of the present invention;

[0066] Figure 7 This is a structure of an electronic device provided by an exemplary embodiment of the present invention. DETAILED DESCRIPTION

[0067] Below, the exemplary embodiments according to the present invention will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments of the present invention, and it should be understood that the present invention is not limited to the exemplary embodiments described here.

[0068] It should be noted that the relative arrangement of components and steps, the numerical expressions and numerical values ​​set forth in these embodiments do not limit the scope of the present invention unless specifically stated otherwise.

[0069] Those skilled in the art can understand that the terms "first" and "second" in the embodiments of the present invention are only used to distinguish different steps, devices or modules, etc., and neither represent any specific technical meaning nor indicate the necessary logical order between them.

[0070] It should also be understood that, in the embodiments of the present invention, “plurality” may refer to two or more than two, and “at least one” may refer to one, two or more than two.

[0071] It should also be understood that any component, data or structure mentioned in the embodiments of the present invention can generally be understood as one or more, unless explicitly limited or otherwise indicated in the context.

[0072] In addition, the term "and / or" in the present invention is only a description of the association relationship of the associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in the present invention generally indicates that the associated objects before and after are in an "or" relationship.

[0073] It should also be understood that the description of the various embodiments of the present invention focuses on the differences between the various embodiments, and the same or similar aspects thereof can be referenced to each other, and for the sake of brevity, they will not be described one by one.

[0074] At the same time, it should be understood that for the convenience of description, the sizes of the various parts shown in the drawings are not drawn according to the actual proportional relationship.

[0075] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the invention, its application, or uses.

[0076] Technologies, methods, and equipment known to ordinary technicians in the relevant art may not be discussed in detail, but where appropriate, the technologies, methods, and equipment should be considered part of the specification.

[0077] It should be noted that like reference numerals and letters refer to similar items in the following figures, and therefore, once an item is defined in one figure, it need not be further discussed in subsequent figures.

[0078] Embodiments of the present invention can be applied to electronic devices such as terminal devices, computer systems, servers, etc., which can operate with many other general or special computing system environments or configurations. Examples of well-known terminal devices, computing systems, environments and / or configurations suitable for use with electronic devices such as terminal devices, computer systems, servers, etc. include, but are not limited to: personal computer systems, server computer systems, thin clients, thick clients, handheld or laptop devices, microprocessor-based systems, set-top boxes, programmable consumer electronics, network personal computers, small computer systems, large computer systems, and distributed cloud computing technology environments including any of the above systems, etc.

[0079] Electronic devices such as terminal devices, computer systems, servers, etc. can be described in the general context of computer system executable instructions (such as program modules) executed by computer systems. Generally, program modules can include routines, programs, object programs, components, logic, data structures, etc., which perform specific tasks or implement specific abstract data types. Computer systems / servers can be implemented in a distributed cloud computing environment, where tasks are performed by remote processing devices linked through a communication network. In a distributed cloud computing environment, program modules can be located on local or remote computing system storage media including storage devices.

[0080] Exemplary Methods

[0081] Figure 1 is a flow chart of a method for predicting power outages in a distribution network based on a digital twin model provided by an exemplary embodiment of the present invention. This embodiment can be applied to electronic devices, such as Figure 1 As shown, a distribution network power outage fault prediction method 100 based on a digital twin model includes the following steps:

[0082] Step 101, collecting parameter data related to power outage faults in the distribution network, wherein the parameter data includes electrical parameters, equipment status parameters, meteorological data, power grid operation information, and historical power outage fault records;

[0083] Step 102, constructing a distribution network digital twin model based on the preprocessed parameter data and the structural information of the distribution network;

[0084] Step 103, using the XGBoost algorithm to train the multi-source time series data of the distribution network twin digital model to generate a power outage fault prediction model;

[0085] Step 104: input the real-time operating parameter data of the distribution network into the power outage fault prediction model for prediction, and output the power outage fault prediction result of the distribution network.

[0086] Specifically, the present invention proposes a power outage analysis and prediction method based on digital twin technology and CNN (convolutional neural network)-LSTM (long short-term memory network)-Attention (attention mechanism) and XGBoost (distributed gradient boosting), aiming to overcome the limitations of the existing technology and enhance the real-time dynamic perception and prediction capabilities of power grid power outage analysis, so as to better serve the intelligent management and optimization of the power grid.

[0087] The working principle of the present invention is as follows: the present invention builds a virtual model of the distribution network system based on digital twin technology, which interacts with the actual distribution network system in real time and reflects the state, structure and operating characteristics of the physical system in real time by collecting multi-source data. On this basis, the CNN-LSTM-Attention model in the deep learning algorithm is used to deeply mine and analyze the power outage fault related data, and the XGBoost algorithm is introduced to further improve the prediction performance, so as to achieve accurate prediction and cause diagnosis of power outage faults.

[0088] refer to Figure 2 As shown, the method provided by the present invention specifically comprises the following steps:

[0089] 1. Data collection and preprocessing:

[0090] 1) Data collection: Build a comprehensive distribution network data collection system to obtain data related to power outages from multiple data sources. These data sources include but are not limited to various sensors installed on transmission lines, substations, and distribution equipment to collect electrical parameters (such as voltage, current, power factor, etc.), equipment status parameters (such as temperature, vibration, insulation performance, etc.); meteorological monitoring systems to provide real-time meteorological data (such as temperature, humidity, wind speed, rainfall, lightning activity, etc.); power grid operation monitoring systems to obtain information such as power grid topology, switch status, load changes, etc.; and historical power outage fault record databases, which contain detailed information on previous power outage events (such as power outage time, location, duration, recovery process, fault cause, etc.), as shown in the following table.

[0091] Through various sensors widely deployed in the distribution network, such as voltage transformers, current transformers, power sensors, etc., the operation parameter data of the distribution network are comprehensively collected, including voltage, current, power factor, active power, reactive power, etc. These data are the basis for building a digital twin model of the distribution network and conducting power outage fault analysis and prediction.

[0092] Table 1 Table of characteristic variables related to power outage

[0093]

[0094] 2) Data preprocessing:

[0095] a. Outlier processing: Use the Z-score method to remove outliers. This method is based on the mean and standard deviation of the data. Data that deviates from the mean by a certain multiple of the standard deviation is considered an outlier and removed to ensure the rationality of the data. For example, in voltage data collection, if the voltage value at a certain moment deviates too much from the normal voltage range, it may be caused by measurement errors or potential faults. The Z-score method can effectively identify and remove such outliers.

[0096] b. Noise removal: Use low-pass filters to remove noise interference. Low-pass filters allow low-frequency signals to pass through, effectively filtering out high-frequency noise, making data smoother and improving data quality. During the current data acquisition process, noise may be generated due to factors such as electromagnetic interference. Low-pass filters can remove this noise, allowing current data to more accurately reflect the actual operating status.

[0097] c. Optimize transmission and computing efficiency: Use data sharding and batch processing technology to optimize data transmission and computing efficiency. Data sharding divides large-scale data into smaller fragments for easy transmission and processing; batch processing technology processes multiple data fragments in a centralized manner, reducing the number of processing times and improving computing efficiency. This is particularly important when processing massive amounts of distribution network operation data, ensuring timely data transmission and processing to meet the real-time requirements of power outage fault analysis and prediction.

[0098] 2. Construction and real-time update of distribution network digital twin model

[0099] According to the physical structure, equipment characteristics and operation principle of the distribution network, combined with the collected grid topology data, equipment parameter information and real-time operation data, a distribution network digital twin model is constructed. The model should accurately describe the connection relationship, electrical characteristics and dynamic behavior between the various components of the distribution network (such as distribution equipment, loads, etc.), and realize high-fidelity mapping of the physical distribution network in virtual space.

[0100] As the distribution network operates in real time, new data is continuously collected, and the digital twin model is dynamically updated using an incremental update algorithm. When the topology of the power grid changes (such as new transmission lines, substation expansion, equipment increase or decrease, etc.), the corresponding nodes and connection relationships are added or modified in the model in a timely manner; when the equipment parameters change (such as resistance changes and capacitance changes caused by equipment aging) or the equipment status is updated (such as equipment status changes after maintenance), the equipment parameters and status information in the model are adjusted accordingly to ensure that the digital twin model is always highly consistent with the actual distribution network and truly reflects its current status and change trends.

[0101] 3. If Figure 3As shown in the figure, power outage fault analysis and prediction based on CNN (convolutional neural network)-LSTM (long short-term memory network)-Attention (attention mechanism)-XGBoost (distributed gradient boosting):

[0102] 1) Feature extraction and data reconstruction (CNN layer, convolutional neural network)

[0103] The multi-source time series data of the distribution network twin digital model is input into the CNN layer, and the data format is (number of samples, time step, number of features). CNN extracts local features of the data by sliding the convolution kernel on the time series. For example, for the time series data of an electrical parameter, the convolution kernel can capture the changing trend and fluctuation characteristics of the data in a local time period. The size of the convolution kernel can be reasonably selected according to the characteristics of the data and computing resources, and is set to k×k (k is usually a smaller value such as 3 or 5). For a feature sequence x and a convolution kernel w, the convolution operation formula is (where i represents the sliding position of the convolution kernel in the time step direction, and j and l represent the row and column positions of the convolution kernel on the two-dimensional feature plane, respectively.) Different local feature patterns are captured by combining multiple convolution kernels and activation functions (such as ReLU).

[0104] Then the pooling operation is performed, using the maximum pooling or average pooling method, and the formula is (M Pooling (x,ps))[i]=max{x[i+j-1,l],j=1,…,ps,l=1,…,ps}(ps is the pooling window size), reducing the data dimension and retaining key information. After convolution and pooling operations, the data dimension changes, and sequence data containing local features is obtained, which is then reconstructed into a format suitable for LSTM processing.

[0105] 2) If Figure 4 and Figure 5 As shown, time series feature learning and global feature acquisition (LSTM layer, long short-term memory network)

[0106] The LSTM layer receives the feature data output by the CNN layer. Its core is to process time series data through memory units and gating mechanisms. The LSTM unit contains an input gate i t 、Forget gate t , Update cell status C t and output gate o t .

[0107] Input Gate i t =σ(W xi ·x t +W hi ·h t-1 +b i)(where x t is the input vector at the current moment, h t-1 is the hidden state vector at the previous moment, W xi and W hi is the weight matrix, b i is the bias term, σ represents the Sigmoid activation function), which controls the update of the cell state.

[0108] Forget Gate t =σ(W xf ·x t +W hf ·h t-1 +b f )(W xf and W hf is the weight matrix, b f is the bias), which determines the degree of forgetting of the cell state at the previous moment.

[0109] Update cell status C t =f t ⊙C t-1 +i t ⊙tanh(W xc ·x t +W hc ·h t-1 +b c )(C t-1 is the cell state at the previous moment, and tanh represents the hyperbolic tangent activation function).

[0110] Output gate o t =σ(W xo ·x t +W ho ·h t-1 +b o )(W xo and W ho is the weight matrix, b o is the bias), which controls the mapping between the cell state and the hidden state.

[0111] The final hidden state h t =o t ⊙tanh(C t ). The LSTM cell output at each time step is equal to the input x t Vectors of the same shape are processed by the LSTM layer, which enhances the feature representation of time series data, extracts the complex features of the data, and provides rich information for the subsequent Attention mechanism and classification layer.

[0112] 3) Attention mechanism weighting and prediction (Attention layer)

[0113] The Attention layer calculates the attention weights of the input features. For the LSTM encoding layer output H (dimension is (S, U), S represents the time step, and U represents the number of hidden units), first calculate the attention weight matrix A, the formula is E = tanh (W attention ·H+b attention ),A=softmax(E),(W attention is the learnable parameter matrix, b attention for bias).

[0114] Then perform a dot product operation on the attention weight matrix A and the LSTM encoding layer output H to obtain the weighted representation H of each feature dimension weighted =A⊙H, making the model focus more on key features.

[0115] 4) Prediction and optimization based on XGBoost (distributed gradient boosting)

[0116] The weighted feature data output by the Attention layer is used as the input of XGBoost. As an integrated machine learning algorithm, XGBoost adds regularization terms to the gradient boosting decision tree to avoid overfitting. Assuming there are N input data and E decision trees, the expression of the decision tree integration model and the objective function is: (in are the true value and predicted value of the nth data respectively; is the e-th decision tree function in the decision tree space F, corresponding to the structure of the e-th tree and the relevant information of the leaf weights; l(·) is the training error function; ω(f e ) is a regularization term to control the complexity).

[0117] The weighted feature data is trained and predicted by the XGBoost algorithm to obtain the final power outage fault prediction result. For multi-classification problems (such as different types of power outages, such as line faults, equipment faults, and meteorological disasters), the number of prediction categories is n, and for the input value x of the jth category, j , output the probability value of the power outage fault belonging to each category, so as to determine the type of power outage fault, and combine the digital twin model to predict the power outage area and evaluate the impact range of the power outage.

[0118] 4. Result evaluation and feedback optimization

[0119] The prediction results are comprehensively compared with the actual power outage fault conditions, and a variety of evaluation indicators are used to measure the model performance. The accuracy calculation formula is: (TP is the true positive, i.e. the number of times a power outage is correctly predicted and actually occurs; FP is the false positive, i.e. the number of times a power outage is incorrectly predicted but does not actually occur), which is used to measure the accuracy of the model prediction; the recall calculation formula is: (F is the false negative, that is, the number of power outages that actually occurred but were not predicted), reflecting the model's coverage of actual power outages; the F1 score calculation formula is: Comprehensively evaluate the performance of the model.

[0120] According to the evaluation results, the feedback mechanism is used to optimize and adjust the model. When it is found that the model has a low accuracy in predicting power outages in specific scenarios or specific types of power outages, the reasons are analyzed in depth. For example, if it is a data problem, further optimize the data collection and preprocessing methods; if it is a model structure or parameter problem, adjust the structure of the CNN-LSTM-Attention-XGBoost model (such as increasing or decreasing the number of convolutional layers and LSTM layers, adjusting the size of the convolution kernel, etc.), optimize the model parameters (such as using a more appropriate initialization method, adjusting the learning rate, etc.), or increase the training data related to the scenario or fault type to improve the generalization ability and prediction accuracy of the model, and continuously improve the model's ability to analyze and predict power outages.

[0121] The method of the present invention is applied to a certain city's distribution network to establish a digital twin model covering the entire city's distribution network. Various types of data are collected in real time through a widely deployed sensor network and data acquisition system. In the actual operation process, multiple power outages were successfully predicted in advance, including power outages caused by transmission line failures due to heavy rain and local power outages caused by transformer aging. Power companies carried out equipment maintenance and line maintenance in a timely manner based on the prediction results, avoiding the occurrence of large-scale power outages, significantly improving the reliability and stability of the power supply in the area, and providing a strong guarantee for the normal operation of the city.

[0122] Therefore, the present invention integrates advanced digital twin technology and deep learning algorithms to construct a highly accurate and real-time dynamically updated digital twin model of the distribution network, which can accurately analyze and predict power outages in advance, and effectively improve the operation management level and power supply reliability of the distribution network system. It can achieve comprehensive, real-time and accurate analysis and prediction of power outages in the distribution network. The close integration of the digital twin model and the actual system makes the monitoring of the distribution network status more accurate and the prediction of power outages more timely and reliable. This helps power companies to formulate response measures in advance, reasonably arrange operation and maintenance resources, quickly restore power supply, effectively reduce the impact of power outages on social production and life, improve the overall operation efficiency and reliability of the distribution network, and ensure the stability and safety of power supply.

[0123] Exemplary Devices

[0124] Figure 6 FIG. 1 is a schematic diagram of a power distribution network outage fault prediction device based on a digital twin model provided by an exemplary embodiment of the present invention. Figure 6 As shown, the device 600 includes:

[0125] The acquisition module 610 is used to collect parameter data related to power outage faults in the distribution network, wherein the parameter data includes electrical parameters, equipment status parameters, meteorological data, power grid operation information and historical power outage fault records;

[0126] A construction module 620 is used to construct a distribution network digital twin model based on the preprocessed parameter data and the structural information of the distribution network;

[0127] A training module 630 is used to train multi-source time series data of the distribution network twin digital model using an XGBoost algorithm to generate a power outage fault prediction model;

[0128] The prediction module 640 is used to input the real-time operating parameter data of the distribution network into the power outage fault prediction model for prediction, and output the power outage fault prediction result of the distribution network.

[0129] Optionally, the preprocessing operation of the parameter data in the construction module 620 includes:

[0130] The Z-score method was used to remove outliers in parameter data;

[0131] A low-pass filter is used to remove noise interference in the parameter data to obtain preprocessed parameter data.

[0132] Optionally, the apparatus 600 further includes:

[0133] The update module is used to dynamically update the distribution network digital twin model based on the continuously collected real-time parameter data of the distribution network using an incremental update algorithm.

[0134] Optionally, the training module 630 includes:

[0135] The local feature extraction submodule is used to input the multi-source time series data of the distribution network twin digital model into the convolutional neural network for feature extraction and obtain the local features of the multi-source time series data;

[0136] The global feature extraction submodule is used to input local features into the long short-term memory network for global feature extraction and obtain the global features of multi-source time series data;

[0137] The weighted submodule is used to input the global features into the attention mechanism layer for attention weighting and output the weighted feature data of multi-source time series;

[0138] The training submodule is used to train and predict the weighted feature data through the XGBoost algorithm to obtain the historical power outage fault prediction results of the distribution network;

[0139] The evaluation submodule is used to compare and evaluate the historical power outage fault prediction results with the historical actual power outage fault conditions to determine the model evaluation results;

[0140] The adjustment submodule is used to optimize and adjust the model using a feedback mechanism according to the model evaluation results to generate a power outage fault prediction model.

[0141] Optionally, the local feature extraction submodule includes:

[0142] The extraction unit extracts the initial local features of multi-source time series data by sliding the convolution kernel on the time series;

[0143] The reconstruction unit is used to perform pooling operations on the initial local features and reconstruct them, and output local features in a format suitable for long short-term memory network processing.

[0144] Optionally, the convolution operation formula is:

[0145]

[0146] In the formula, x is the feature sequence of multi-source time series data; w is the convolution kernel; i represents the sliding position of the convolution kernel in the time step direction, j and l represent the row and column positions of the convolution kernel on the two-dimensional feature plane respectively;

[0147] The pooling operation formula is:

[0148] (M Pooling (x,ps))[i]=max{x[i+j-1,l],j=1,…,ps,l=1,…,ps}

[0149] Where ps is the pooling window size.

[0150] Optionally, the LSTM input gate i t 、Forget gate t , Update cell status C t and output gate o t ,in,

[0151] Input Gate i t =σ(W xi ·x t +W hi ·h t-1 +b i ), controls the update of cell state, where x t is the input vector at the current moment, h t-1is the hidden state vector at the previous moment, W xi and W hi is the weight matrix, b i is the bias term, σ represents the Sigmoid activation function;

[0152] Forget Gate t =σ(W xf ·x t +W hf ·h t-1 +b f ), determines the degree of forgetting of the cell state at the previous moment, where W xf and W hf is the weight matrix, b f is bias;

[0153] Update cell status C t =f t ⊙C t-1 +i t ⊙tanh(W xc ·x t +W hc ·h t-1 +b c ), where C t-1 is the cell state at the previous moment, tanh represents the hyperbolic tangent activation function;

[0154] Output gate o t =σ(W xo ·x t )+W ho ·h t-1 +b o ), controls the mapping between cell state and hidden state, where W xo and W ho is the weight matrix, b o is bias;

[0155] The final hidden state of the LSTM network is h t =o t ⊙tanh(C t ).

[0156] Optionally, the weight matrix A of the attention mechanism layer is calculated as:

[0157] A=softmax(E)

[0158] Where E = tanh(W attention ·H+b attention ), W attention is the learnable parameter matrix, b attention is the bias, and H is the global feature output by the long short-term memory network.

[0159] Optionally, the decision tree ensemble model of the XGBoost algorithm is:

[0160]

[0161] The objective function is:

[0162]

[0163] In the formula, are the true value and predicted value of the nth data respectively; is the e-th decision tree function in the decision tree space F, corresponding to the structure of the e-th tree and the relevant information of the leaf weights; l(·) is the training error function; ω(f e ) is the regular term for controlling complexity, N is the number of input data, and E is the number of decision trees.

[0164] Optionally, an evaluation submodule includes:

[0165] The calculation submodule is used to calculate the accuracy and recall rate of the model based on the historical power outage fault prediction results and the historical actual power outage fault conditions;

[0166] Determine the submodule, which is used to calculate the F1 score based on the precision and recall rate and determine the model evaluation result.

[0167] Optionally, the F1 score is calculated as:

[0168]

[0169] in, TP is the true positive, which is the number of times the power outage is correctly predicted and actually occurs; FP is the false positive, which is the number of times the power outage is incorrectly predicted but does not actually occur. FN is the false negative, which is the number of times a power outage actually occurred but was not predicted.

[0170] Exemplary Electronic Devices

[0171] Figure 7 This is a structure of an electronic device provided by an exemplary embodiment of the present invention. Figure 7 As shown, the electronic device 70 includes one or more processors 71 and a memory 72 .

[0172] The processor 71 may be a central processing unit (CPU) or other forms of processing units having data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device to perform desired functions.

[0173] The memory 72 may include one or more computer program products, and the computer program product may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, random access memory (RAM) and / or cache memory (cache), etc. The non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and the processor 71 may run the program instructions to implement the methods of the software programs of the various embodiments of the present invention described above and / or other desired functions. In one example, the electronic device may also include: an input device 73 and an output device 74, which are interconnected via a bus system and / or other forms of connection mechanisms (not shown).

[0174] In addition, the input device 73 may also include, for example, a keyboard, a mouse, and the like.

[0175] The output device 74 can output various information to the outside, and can include, for example, a display, a speaker, a printer, a communication network and a remote output device connected thereto.

[0176] Of course, to simplify, Figure 7 Only some of the components related to the present invention in the electronic device are shown, and components such as a bus, an input / output interface, etc. are omitted. In addition, the electronic device may further include any other appropriate components according to specific application conditions.

[0177] Exemplary computer program products and computer-readable storage media

[0178] In addition to the above-mentioned methods and devices, an embodiment of the present invention may also be a computer program product, which includes computer program instructions, which, when executed by a processor, enable the processor to execute the steps of the method according to various embodiments of the present invention described in the above-mentioned "Exemplary Method" section of this specification.

[0179] The computer program product may be written in any combination of one or more programming languages ​​to write program code for performing the operations of the embodiments of the present invention, including object-oriented programming languages ​​such as Java, C++, etc., and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as a separate software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0180] In addition, an embodiment of the present invention may also be a computer-readable storage medium having computer program instructions stored thereon, which, when executed by a processor, enable the processor to execute the steps of the method according to various embodiments of the present invention described in the above “Exemplary Method” section of this specification.

[0181] The computer readable storage medium can adopt any combination of one or more readable media. The readable medium can be a readable signal medium or a readable storage medium. The readable storage medium can include, for example, but is not limited to, a system, system or device of electricity, magnetism, light, electromagnetic, infrared, or semiconductor, or any combination of the above. More specific examples (non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0182] The basic principle of the present invention is described above in conjunction with specific embodiments. However, it should be pointed out that the advantages, strengths, effects, etc. mentioned in the present invention are only examples and not limitations, and it cannot be considered that these advantages, strengths, effects, etc. must be possessed by each embodiment of the present invention. In addition, the specific details disclosed above are only for the purpose of illustration and facilitation of understanding, rather than limitation, and the above details do not limit the present invention to being implemented by adopting the above specific details.

[0183] Each embodiment in this specification is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the system embodiment, since it basically corresponds to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0184] The block diagrams of the devices, systems, equipment, and systems involved in the present invention are only illustrative examples and are not intended to require or imply that they must be connected, arranged, and configured in the manner shown in the block diagram. As will be appreciated by those skilled in the art, these devices, systems, equipment, and systems can be connected, arranged, and configured in any manner. Words such as "including", "comprising", "having", etc. are open words, referring to "including but not limited to", and can be used interchangeably with them. The words "or" and "and" used here refer to the words "and / or" and can be used interchangeably with them, unless the context clearly indicates otherwise. The word "such as" used here refers to the phrase "such as but not limited to", and can be used interchangeably with it.

[0185] The method and system of the present invention may be implemented in many ways. For example, the method and system of the present invention may be implemented by software, hardware, firmware or any combination of software, hardware, firmware. The above order of steps for the method is only for illustration, and the steps of the method of the present invention are not limited to the order specifically described above, unless otherwise specifically stated. In addition, in some embodiments, the present invention may also be implemented as a program recorded in a recording medium, which includes machine-readable instructions for implementing the method according to the present invention. Thus, the present invention also covers a recording medium storing a program for executing the method according to the present invention.

[0186] It should also be noted that in the system, device and method of the present invention, each component or each step can be decomposed and / or recombined. These decompositions and / or recombinations should be regarded as equivalent schemes of the present invention. The above description of the disclosed aspects is provided to enable any technician in the field to make or use the present invention. Various modifications to these aspects are very obvious to those skilled in the art, and the general principles defined here can be applied to other aspects without departing from the scope of the present invention. Therefore, the present invention is not intended to be limited to the aspects shown here, but in accordance with the widest range consistent with the principles and novel features disclosed here.

[0187] The above description has been given for the purpose of illustration and description. In addition, this description is not intended to limit the embodiments of the present invention to the forms disclosed herein. Although a number of example aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, changes, additions and sub-combinations thereof.

Claims

1. A distribution network power outage fault prediction method based on a digital twin model, characterized in that: include: Collecting parameter data related to power outage faults in the distribution network, wherein the parameter data includes electrical parameters, equipment status parameters, meteorological data, power grid operation information, and historical power outage fault records; Constructing a distribution network digital twin model according to the preprocessed parameter data and the structural information of the distribution network; The XGBoost algorithm is used to train the multi-source time series data of the distribution network twin digital model to generate a power outage fault prediction model; The real-time operating parameter data of the distribution network is input into the power outage fault prediction model for prediction, and the power outage fault prediction result of the distribution network is output.

2. The method according to claim 1, characterized in that The preprocessing operation of the parameter data includes: Using the Z-score method to remove outliers in the parameter data; A low-pass filter is used to remove noise interference in the parameter data to obtain preprocessed parameter data.

3. The method according to claim 1, characterized in that Also includes: An incremental update algorithm is used to dynamically update the distribution network digital twin model based on the continuously collected real-time parameter data of the distribution network.

4. The method according to claim 1, characterized in that: The XGBoost algorithm is used to train the multi-source time series data of the distribution network twin digital model to generate a power outage fault prediction model, including: Inputting the multi-source time series data of the distribution network twin digital model into a convolutional neural network for feature extraction to obtain local features of the multi-source time series data; Inputting the local features into a long short-term memory network to extract global features, thereby obtaining global features of the multi-source time series data; Inputting the global features into the attention mechanism layer for attention weighting, and outputting weighted feature data of the multi-source time series; The weighted feature data is trained and predicted by an XGBoost algorithm to obtain a historical power outage fault prediction result of the distribution network; Compare and evaluate the historical power outage fault prediction results with the historical actual power outage fault conditions to determine the model evaluation results; The model is optimized and adjusted using a feedback mechanism according to the model evaluation result to generate the power outage fault prediction model.

5. The method according to claim 1, characterized in that Inputting the multi-source time series data of the distribution network twin digital model into a convolutional neural network for feature extraction to obtain local features of the multi-source time series data includes: Extracting initial local features of the multi-source time series data by sliding a convolution kernel on the time series; The initial local features are pooled and reconstructed to output the local features in a format suitable for processing by the long short-term memory network.

6. The method according to claim 5, characterized in that The convolution operation formula is: In the formula, x is the feature sequence of multi-source time series data; w is the convolution kernel; i represents the sliding position of the convolution kernel in the time step direction, j and l represent the row and column positions of the convolution kernel on the two-dimensional feature plane respectively; The pooling operation formula is: (M Pooling (x,ps))[i]=max{x[i+j-1,l],j=1,…,ps,l=1,…,ps} Where ps is the pooling window size.

7. The method according to claim 4, characterized in that The long short-term memory network input gate i t 、Forget gate t , Update cell status C t and output gate o t ,in, Input Gate i t =σ(W xi ·x t +W hi ·h t-1 +b i ), controls the update of cell state, where x t is the input vector at the current moment, h t-1 is the hidden state vector at the previous moment, W xi and W hi is the weight matrix, b i is the bias term, σ represents the Sigmoid activation function; Forget Gate t =σ(W xf ·x t +W hf ·h t-1 +b f ), determines the degree of forgetting of the cell state at the previous moment, where W xf and W hf is the weight matrix, b f is bias; Update cell status C t =f t ⊙C t-1 +i t ⊙tanh(W xc ·x t +W hc ·h t-1 +b c ), where C t-1 is the cell state at the previous moment, tanh represents the hyperbolic tangent activation function; Output gate o t =σ(W xo ·x t +W ho ·h t-1 +b o ), controls the mapping between cell state and hidden state, where W xo and W ho is the weight matrix, b o is bias; The final hidden state of the LSTM network is h t =o t ⊙tanh(C t ).

8. The method according to claim 4, characterized in that The calculation formula of the weight matrix A of the attention mechanism layer is: A=softmax(E) Where E = tanh(W attention ·H+b attention ), W attention is the learnable parameter matrix, b attention is the bias, and H is the global feature output by the long short-term memory network.

9. The method according to claim 4, characterized in that The decision tree ensemble model of the XGBoost algorithm is: The objective function is: In the formula, are the true value and predicted value of the nth data respectively; is the e-th decision tree function in the decision tree space F, corresponding to the structure of the e-th tree and the relevant information of the leaf weights; l(·) is the training error function; ω(f e ) is the regular term for controlling complexity, N is the number of input data, and E is the number of decision trees.

10. The method according to claim 4, characterized in that Compare and evaluate the historical power outage fault prediction results with the historical actual power outage fault conditions to determine the model evaluation results, including: Calculate the accuracy and recall of the model based on the historical power outage fault prediction results and the historical actual power outage fault conditions; The F1 score is calculated based on the precision and the recall to determine the model evaluation result.

11. The method according to claim 10, characterized in that The calculation formula of the F1 score is: in, TP is the true positive, that is, the number of times the power outage fault is correctly predicted and actually occurs; FP is the false positive, that is, the number of times the power outage fault is incorrectly predicted but does not actually occur; FN is the false negative, which is the number of times a power outage actually occurred but was not predicted.

12. A distribution network power outage fault prediction device based on a digital twin model, characterized in that: include: A collection module is used to collect parameter data related to power outage faults in the distribution network, wherein the parameter data includes electrical parameters, equipment status parameters, meteorological data, power grid operation information and historical power outage fault records; A construction module, used to construct a distribution network digital twin model based on the preprocessed parameter data and the structural information of the distribution network; A training module, used to train the multi-source time series data of the distribution network twin digital model using an XGBoost algorithm to generate a power outage fault prediction model; The prediction module is used to input the real-time operating parameter data of the distribution network into the power outage fault prediction model for prediction, and output the power outage fault prediction result of the distribution network.

13. The device according to claim 12, characterized in that Also includes: An update module is used to dynamically update the distribution network digital twin model based on the continuously collected real-time parameter data of the distribution network using an incremental update algorithm.

14. The device according to claim 12, characterized in that Training modules include: A local feature extraction submodule, used for inputting the multi-source time series data of the distribution network twin digital model into a convolutional neural network for feature extraction, and obtaining local features of the multi-source time series data; A global feature extraction submodule is used to input the local features into a long short-term memory network to extract global features, so as to obtain the global features of the multi-source time series data; A weighting submodule, used for inputting the global features into the attention mechanism layer for attention weighting, and outputting weighted feature data of the multi-source time series; A training submodule, used for training and predicting the weighted feature data by using an XGBoost algorithm to obtain a historical power outage fault prediction result of the distribution network; An evaluation submodule is used to compare and evaluate the historical power outage fault prediction results with the historical actual power outage fault conditions to determine the model evaluation results; The adjustment submodule is used to optimize and adjust the model using a feedback mechanism according to the model evaluation result to generate the power outage fault prediction model.

15. The device according to claim 12, characterized in that Local feature extraction submodule, including: An extraction unit, used for extracting initial local features of the multi-source time series data by sliding a convolution kernel on the time series and performing a convolution operation; The reconstruction unit is used to perform a pooling operation on the initial local features and reconstruct the local features to output the local features in a format suitable for processing by the long short-term memory network.

16. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and the computer program is used to execute the method according to any one of claims 1 to 11.

17. An electronic device, characterized in that: The electronic device comprises: processor; a memory for storing instructions executable by the processor; The processor is used to read the executable instructions from the memory and execute the instructions to implement the method described in any one of claims 1-11.

Images