Intelligent power distribution network fault prediction and self-healing method based on big data analysis and edge calculation

Abstract

The invention discloses an intelligent power distribution network fault prediction and self-healing method based on big data analysis and edge computing, which comprises an edge computing node, a cloud server and a communication module, and is characterized in that the edge computing node and the cloud server are respectively in communication connection with the communication module; the edge computing nodes are used for collecting current, voltage and temperature data in real time, the edge computing nodes are deployed at section switches of a power distribution network, and the cloud server is used for storing historical data and training a fault prediction model; and the communication module is used for realizing data interaction between the edge node and the cloud, and supports 5G / optical fibers. The method has the advantages that the real-time performance is improved, and the fault response time is shortened from the hour level to the second level. And multi-source data fusion enables the fault identification accuracy to be greater than or equal to 95%. And privacy and cost optimization: the cloud data transmission amount is reduced by more than 50% through edge calculation, and user data is protected by differential privacy. And adaptability is enhanced, and a complex power distribution network topology containing distributed energy is supported.

Description

Technical Field

[0001] The present invention belongs to the technical field of smart grids, and particularly relates to a method for fault prediction and self-healing of a smart distribution network based on big data analysis and edge computing. Background Art

[0002] A smart distribution network is a distribution system that integrates traditional and cutting-edge distribution engineering technologies, advanced sensing and measurement and control technologies, and modern computer and communication technologies. It is safer, more reliable, of better quality, and more efficient, and supports the large-scale access of distributed power sources.

[0003] Chinese invention patent CN 111429034 A discloses a method for fault prediction of a distribution network, including the first step: data preprocessing part, including analysis of distribution network fault influencing factors and sample screening based on a particle swarm algorithm; the second step: feature variable selection part, including preliminary feature selection of each relevant variable, and then re-screening using a specific selection algorithm to form an optimal feature variable set; the third step: construction of a distribution network fault prediction model based on fault level division; dividing the faults into levels, optimizing the parameters of the support vector machine using an improved particle swarm optimization algorithm, and then constructing a distribution network fault prediction model, analyzing various data of the distribution network based on the constructed model, and outputting anomalies and fault predictions. This solution realizes accurate fault prediction of the distribution network by optimizing data to build an accurate distribution network fault prediction model.

[0004] The above solution and the traditional distribution network fault handling that rely on a centralized master station system both have the following defects: 1. Poor real-time performance: Data needs to be uploaded to the cloud for centralized processing, with high communication latency and a fault response time up to the hour level. 2. Privacy and bandwidth pressure: Uploading a large amount of raw data to the cloud is prone to leaking user privacy and occupies high bandwidth resources. 3. Low prediction accuracy: Single electrical quantity feature analysis cannot adapt to complex fault scenarios, and the fault misjudgment rate is high. 4. Rigid self-healing strategy: Relying on fixed protection settings, it cannot dynamically adapt to network topology changes (such as the access of distributed energy).

[0005] Therefore, it is urgent to develop a method for fault prediction and self-healing of a smart distribution network based on big data analysis and edge computing to solve the problems raised in the above background art. Summary of the Invention

[0006] Aiming at the problems raised in the above background art, the object of the present invention is to provide a method for fault prediction and self-healing of a smart distribution network based on big data analysis and edge computing, which realizes real-time prediction, location, isolation, and recovery of faults through the cooperation of distributed edge computing nodes and the cloud, and significantly improves the reliability and self-healing ability of the distribution network.

[0007] To achieve the above technical object, the technical solution adopted by the present invention is as follows:

[0008] A smart distribution network fault prediction and self-healing system based on big data analysis and edge computing, including edge computing nodes, a cloud server, and a communication module. The edge computing nodes and the cloud server are respectively communicatively connected to the communication module;

[0009] The edge computing nodes are used to collect current, voltage, and temperature data in real time. The edge computing nodes are deployed at each sectional switch of the distribution network. The cloud server is used to store historical data and train a fault prediction model; the communication module is used to realize data interaction between the edge nodes and the cloud, and the communication module supports 5G / fiber optic.

[0010] Further defined, the edge computing node is built-in with a data preprocessing module and a fault prediction module;

[0011] The data preprocessing module uses wavelet transform to filter out noise and extract fault feature quantities;

[0012] The fault prediction module predicts the line fault probability based on the LSTM neural network. The formula is:

[0013] Pf(t)=σ(W·[h t-1 ,xt]+b)

[0014] Where W is the weight matrix, obtained through training and learning; h t-1 Is the hidden state of the previous moment of LSTM, storing historical time series information; xt is the current input feature, including real-time data of current, voltage, and temperature; b is the bias term, used to adjust the model output; σ is the Sigmoid activation function, mapping the output to the [0,1] interval, representing the fault probability.

[0015] Further defined, the fault prediction module fuses the local outlier factor algorithm to calculate the line anomaly index. The formula is:

[0016]

[0017] Where N k (p) is the k-nearest neighbor set of data point p, and lrd(p) is the local reachability density, used to identify line abnormal fluctuations; if LOF k (p)>1, it indicates that the local density is lower than its neighbors and belongs to an outlier point. The outlier points include current mutations.

[0018] Further defined, the cloud server uses the random forest algorithm to generate global fault feature weights and dynamically update the prediction model parameters of the edge computing nodes.

[0019] A smart distribution network fault prediction and self-healing method based on big data analysis and edge computing:

[0020] Step 1: After the edge computing node detects the fault characteristics, it triggers the local circuit breaker to trip;

[0021] Step 2: Calculate the optimal restoration path based on the Dijkstra algorithm. The formula is:

[0022]

[0023] where w ij is the line weight, and the line weight includes the load rate and length. x ij is the path selection variable, and the variable is 0 or 1, indicating whether the edge (i,j) is selected. s,t: the start and end points of the path. The start point of the path is upstream of the fault point, and the end point is the tie switch or backup power supply.

[0024] Further defined, in the scenario of distributed energy access, the self-healing strategy is dynamically adjusted through the Q-Learning algorithm, and the reward function is designed as:

[0025] R = α·recovery speed + β·power supply reliability - γ·operation cost

[0026] where the recovery speed is in seconds; the power supply reliability is the load rate of the power supply area after recovery; the operation cost is the number of switch operations.

[0027] Further defined, the communication module supports the identification of illegal communication links based on signal strength clustering, and such a design can prevent false base station attacks.

[0028] Further defined, the edge computing node performs differential privacy processing on sensitive data before uploading it to the cloud server. The calculation method of the added noise amount is:

[0029]

[0030] where λ is the query sensitivity, defined as the maximum difference in the query results of adjacent data sets; Δf is the privacy budget, and the privacy budget takes values from 0.1 to 1. The smaller the value, the stronger the privacy protection; the sensitive data includes the user load curve.

[0031] Further defined, it also includes closed-loop lines and open-loop lines; in the closed-loop lines, the tie switch is preferentially enabled to realize load transfer; in the open-loop lines, power supply is restored by reconstructing the topology.

[0032] A self-healing effect evaluation model also includes using SAIDI and SAIFI as core indicators. The calculation formula is:

[0033]

[0034] Among them, the SAIDI is the System Average Interruption Duration, and the SAIFI is the System Average Interruption Frequency.

[0035] The working logic of the present invention:

[0036] The present invention collects current and voltage signals through edge computing nodes, extracts features through wavelet denoising; predicts the fault probability through the LSTM model, and identifies abnormal lines through the LOF algorithm; generates the optimal restoration path through the Dijkstra algorithm, and the Q-Learning dynamic optimization strategy; processes user data through differential privacy technology, and quantifies the self-healing effect by SAIDI / SAIFI.

[0037] The beneficial effects of the present invention:

[0038] Improved real-time performance: The fault response time is shortened from the hour level to the second level.

[0039] High prediction accuracy: The fusion of multi-source data makes the fault identification accuracy ≥ 95%.

[0040] Privacy and cost optimization: Edge computing reduces the cloud data transmission volume by more than 50%, and differential privacy protects user data.

[0041] Enhanced adaptability: Supports complex distribution network topologies with distributed energy sources. Description of the drawings

[0042] The present invention can be further illustrated by the non-limiting embodiments given in the drawings;

[0043] Figure 1 It is the system module diagram of the embodiment of the intelligent distribution network fault prediction and self-healing method based on big data analysis and edge computing of the present invention;

[0044] Figure 2 It is the step flow diagram of the embodiment of the intelligent distribution network fault prediction and self-healing method based on big data analysis and edge computing of the present invention;

[0045] Figure 3 It is the module connection diagram of the embodiment of the intelligent distribution network fault prediction and self-healing method based on big data analysis and edge computing of the present invention.

[0046] The main component symbols are explained as follows: edge computing node 101, cloud server 102, communication module 103, data preprocessing module 104, fault prediction module 105. Specific implementation manners

[0047] To enable those skilled in the art to better understand the present invention, the technical solution of the present invention will be further described below in conjunction with the accompanying drawings and embodiments. The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0048] It should be noted that all directional indications (such as up, down, left, right, front, back...) in the embodiments of the present invention are only used to explain the relative position relationship and movement conditions between components in a specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication will also change accordingly.

[0049] In addition, the descriptions involving "first", "second", etc. in the present invention are only for descriptive purposes, and cannot be understood as indicating or implying their relative importance or implicitly indicating the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include at least one of such features. In addition, the technical solutions between various embodiments can be combined with each other, but it must be based on the ability of those of ordinary skill in the art to implement. When the combination of technical solutions conflicts with each other or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0050] As Figures 1 - 3 shown, the intelligent distribution network fault prediction and self-healing system based on big data analysis and edge computing of the present invention includes an edge computing node 101, a cloud server 102, and a communication module 103. The edge computing node 101 and the cloud server 102 are respectively communicatively connected to the communication module 103;

[0051] The edge computing node 101 is used to collect current, voltage, and temperature data in real time. The edge computing node 101 is deployed at each sectional switch of the distribution network. The cloud server 102 is used to store historical data and train a fault prediction model; the communication module 103 is used to realize data interaction between the edge node and the cloud, and the communication module 103 supports 5G / fiber optic.

[0052] In the actual application of this embodiment, the edge computing node 101 is internally provided with a data preprocessing module 104 and a fault prediction module 105;

[0053] The data preprocessing module 104 uses wavelet transform to filter out noise and extract fault feature quantities;

[0054] The fault prediction module 105 predicts the line fault probability based on the LSTM neural network, and the formula is:

[0055] Pf(t) = σ(W · [h t-1 , xt] + b)

[0056] Where, W is the weight matrix, which is obtained through training and learning; h t-1 is the hidden state of the previous moment of the LSTM, which stores historical time series information; xt is the current input feature, including real-time data of current, voltage, and temperature; b is the bias term, which is used to adjust the model output; σ is the Sigmoid activation function, which maps the output to the interval [0, 1], representing the fault probability.

[0057] In the actual application of this embodiment, the fault prediction module 105 fuses the local outlier factor algorithm to calculate the line anomaly index, and the formula is:

[0058]

[0059] Where, N k (p) is the set of k nearest neighbors of the data point p, and lrd(p) is the local reachability density, which is used to identify the abnormal fluctuations of the line; if LOF k (p) > 1, it indicates that the local density is lower than its neighbors and belongs to an abnormal point, and the abnormal points include current mutations.

[0060] In the actual application of this embodiment, the cloud server 102 uses the random forest algorithm to generate the global fault feature weights and dynamically updates the prediction model parameters of the edge computing node 101.

[0061] Intelligent distribution network fault prediction and self-healing method based on big data analysis and edge computing:

[0062] Step 1: After the edge computing node 101 detects the fault feature, it triggers the local circuit breaker to trip;

[0063] Step 2: Calculate the optimal restoration path based on the Dijkstra algorithm, and the formula is:

[0064]

[0065] Where, the w ij is the line weight, and the line weight includes the load rate and length, and the x ij is the path selection variable, and the variable is 0 or 1, indicating whether the edge (i, j) is selected; the s, t: the starting point and the ending point of the path, the starting point of the path is upstream of the fault point, and the ending point is the tie switch or the backup power supply.

[0066] In the practical application of this embodiment, in the distributed energy access scenario, the self-healing strategy is dynamically adjusted through the Q-Learning algorithm, and the reward function is designed as follows:

[0067] R = α · recovery speed + β · power supply reliability - γ · operation cost

[0068] Among them, the recovery speed is in seconds; the power supply reliability is the load rate of the power supply area after recovery; the operation cost is the number of switch operations.

[0069] In the practical application of this embodiment, the communication module supports the identification of illegal communication links based on signal strength clustering. Such a design can prevent pseudo base station attacks.

[0070] In the practical application of this embodiment, the edge computing node 101 performs differential privacy processing on sensitive data and then uploads it to the cloud server 101. The calculation method of the added noise amount is as follows:

[0071]

[0072] Among them, λ is the query sensitivity, defined as the maximum difference in query results of adjacent data sets; Δf is the privacy budget, and the privacy budget usually takes a value of 0.1 - 1. The smaller the value, the stronger the privacy protection; the sensitive data includes the user load curve.

[0073] In the practical application of this embodiment, it also includes a closed-loop line and an open-loop line; in the closed-loop line, the tie switch is preferentially enabled to realize load transfer; in the open-loop line, power supply is restored by reconstructing the topology.

[0074] A self-healing effect evaluation model also includes using SAIDI and SAIFI as core indicators, and the calculation formula is as follows:

[0075]

[0076] Among them, SAIDI is the system average interruption duration, and SAIFI is the system average interruption frequency.

[0077] Embodiment 1: Self-healing of single-phase grounding faults in urban distribution networks

[0078] Scenario: A single-phase grounding fault occurred in a 10 kV line due to cable aging.

[0079] Step 1: The edge computing node 101 detects a sudden increase in zero-sequence current and the LOF index exceeds the threshold, triggering an early warning.

[0080] Step 2: The cloud server 102 issues the latest model parameters, and the LSTM predicts that the fault probability reaches 92%.

[0081] Step 3: The control center starts the Dijkstra algorithm, isolates the faulty section and switches to the tie line for power supply, and the recovery time ≤ 3 seconds.

[0082] Specifically, the LSTM prediction formula in Step 2 is: Pf(t) = σ(W · [h t-1 , xt] + b);

[0083] Input features: zero-sequence current xt = 15 A (normal range ≤ 5 A), voltage fluctuation ΔV = 10%, temperature T = 65 °C.

[0084] Model parameters: W = [0.2, -0.5, 0.3], ht-1 = 0.6, b = -0.1.

[0085] The calculation process is: Pf(t) = σ(0.2 × 15 + (-0.5) × 10 + 0.3 × 65 - 0.1) = σ(3 - 5 + 19.5 - 0.1) = σ(17.4) ≈ 0.999, and the actual output probability after regularization is 92%.

[0086] Example 2: Tree fault short circuit in rural distribution network

[0087] Scenario: A typhoon causes a tree branch to touch the overhead line, resulting in a short circuit.

[0088] Step 1: The edge computing node 101 identifies overheating of the line through a temperature sensor and triggers the circuit breaker to trip.

[0089] Step 2: Select the recovery path with the minimum operating cost based on Q-Learning (avoiding the faulty branch line).

[0090] Specifically, the Q-Learning reward calculation in Step 2 is as follows: R = α · recovery speed + β · power supply reliability - γ · operating cost, where α is 0.6, β is 0.3, and γ is 0.1;

[0091] Recovery speed = 3 seconds (score 0.6 × 3 = 1.8), power supply reliability = restoring 95% of the load (score 0.3 × 95 = 28.5), operating cost = 2 switch operations (penalty 0.1 × 2 = 0.2).

[0092] Result: Total reward R = 1.8 + 28.5 - 0.2 = 30.1, select the path with the highest reward.

[0093] Example 3: Power supply guarantee in privacy-sensitive areas

[0094] Scenario: The distribution network of a military base needs to prevent data leakage.

[0095] Step 1: The edge computing node 101 adds differential privacy noise to the load data and then uploads it.

[0096] Step 2: The cloud server 102 only receives the de-sensitized data (x′) and completes the global analysis.

[0097] Specifically, the calculation process of differential privacy processing in Step 1 is as follows:

[0098] Among them, the original load data x = 100 kW, the sensitivity Δf = 1 kW, and the privacy budget λ = 0.5.

[0099] Noise volume For example, noise = 3 kW, and the de-sensitized data x′ = 100 + 3 = 103 kW.

[0100] Example 4: Large-scale disaster emergency response

[0101] Scenario: An earthquake causes power grid paralysis in multiple regions.

[0102] Step 1: The edge computing node 101 makes autonomous decisions and gives priority to restoring power supply to hospitals and communication base stations.

[0103] Step 2: The cloud server 102 coordinates multiple edge computing nodes 101 to form a microgrid to achieve black start.

[0104] Since

[0105] Traditional solution: Annual power outage duration = 2 hours, SAIDI = 2 / 10,000 = 0.0002 hours / household.

[0106] Solution of this embodiment: Power outage duration = 0.5 hours, SAIDI = 0.5 / 10,000 = 0.00005 hours / household

[0107] Compared with the traditional solution, the power outage duration of this embodiment is reduced by 75%.

[0108] The above embodiments only exemplarily illustrate the principles and effects of the present invention, rather than limiting the present invention. Any person familiar with this technology can modify or change the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or changes completed by those with ordinary knowledge in the technical field without departing from the spirit and technical ideas disclosed by the present invention should still be covered by the claims of the present invention.

Claims

1. A system for intelligent distribution network fault prediction and self-healing based on big data analysis and edge computing, characterized in that: It includes an edge computing node (101), a cloud server (102) and a communication module (103), and the edge computing node (101) and the cloud server (102) are respectively communicatively connected to the communication module (103); The edge computing node (101) is used to collect current, voltage and temperature data in real time. The edge computing node (101) is deployed at each sectional switch of the distribution network. The cloud server (102) is used to store historical data and train a fault prediction model; the communication module (103) is used to realize data interaction between the edge node and the cloud, and the communication module (103) supports 5G / fiber optic.

2. The system for intelligent distribution network fault prediction and self-healing based on big data analysis and edge computing according to claim 1, characterized in that: The edge computing node (101) is built-in with a data preprocessing module (104) and a fault prediction module (105); The data preprocessing module (104) uses wavelet transform to filter out noise and extract fault feature quantities; The fault prediction module (105) predicts the line fault probability based on the LSTM neural network, and the formula is: Pf(t) = σ(W·[h t-1 , xt] + b) where, W is the weight matrix, obtained through training and learning; h t-1 is the hidden state of the previous moment of the LSTM, storing historical time series information; xt is the current input feature, including real-time data of current, voltage, and temperature; b is the bias term, used to adjust the model output; σ is the Sigmoid activation function, mapping the output to the interval [0,1], representing the fault probability.

3. The system for intelligent distribution network fault prediction and self-healing based on big data analysis and edge computing according to claim 2, characterized in that: The fault prediction module (105) fuses the local outlier factor algorithm to calculate the line anomaly index, and the formula is: Among them, N k (p) is the set of k nearest neighbors of the data point p, and lrd(p) is the local reachability density, which is used to identify abnormal fluctuations in the line; if LOF k (p) > 1, it indicates that the local density is lower than that of its neighbors and belongs to an outlier. The outlier includes current mutation.

4. The system for intelligent distribution network fault prediction and self-healing based on big data analysis and edge computing according to claim 1, characterized in that: The cloud server (102) uses the random forest algorithm to generate global fault feature weights and dynamically updates the prediction model parameters of the edge computing node (101).

5. A smart distribution network fault prediction and self-healing method based on big data analysis and edge computing, characterized in that: Step 1: After the edge computing node (101) detects a fault feature, it triggers the local circuit breaker to trip; Step 2: Calculate the optimal restoration path based on the Dijkstra algorithm, and the formula is: Among them, the w ij is the line weight, and the line weight includes the load factor and the length. The x ij is the path selection variable, and the variable is 0 or 1, indicating whether the edge (i, j) is selected; s and t are the start and end points of the path, the start point of the path is upstream of the fault point, and the end point is the tie switch or the backup power supply.

6. The intelligent distribution network fault prediction and self-healing method based on big data analysis and edge computing according to claim 5, characterized in that: In the distributed energy access scenario, the self-healing strategy is dynamically adjusted through the Q-Learning algorithm, and the reward function is designed as: R = α·recovery speed + β·power supply reliability - γ·operation cost Among them, the recovery speed is in seconds; the power supply reliability is the load rate of the power supply area after restoration; the operation cost is the number of switch operations.

7. The intelligent distribution network fault prediction and self-healing method based on big data analysis and edge computing according to claim 5, characterized in that: The communication module supports the identification of illegal communication links based on signal strength clustering.

8. The intelligent distribution network fault prediction and self-healing method based on big data analysis and edge computing according to claim 5, characterized in that: The edge computing node (101) performs differential privacy processing on sensitive data and then uploads it to the cloud server (101). The calculation method of the added noise amount is: Among them, λ is the query sensitivity, defined as the maximum difference in the query results of adjacent data sets; Δf is the privacy budget, and the privacy budget takes 0.1 to 1, and the smaller the value, the stronger the privacy protection; the sensitive data includes the user load curve.

9. The intelligent distribution network fault prediction and self-healing method based on big data analysis and edge computing according to claim 5, characterized in that: It also includes a closed-loop line and an open-loop line; in the closed-loop line, the tie switch is preferentially enabled to realize load transfer; in the open-loop line, the power supply is restored by reconstructing the topology.

10. A self-healing effect evaluation model, characterized in that: It also includes using SAIDI and SAIFI as core indicators, and the calculation formula is: Among them, SAIDI is the system average interruption duration, and SAIFI is the system average interruption frequency.

Images