A CNN-LSTM-based fault location method for island microgrids

Abstract

The application discloses a CNN-LSTM-based island micro-grid fault positioning method, comprising the following steps: constructing an island micro-grid fault positioning framework; constructing an island micro-grid muPMU optimization configuration model; solving the island micro-grid muPMU optimization configuration model; analyzing electrical variation before and after island micro-grid fault; constructing a CNN-LSTM model; and constructing a CNN-LSTM input and output. The fault positioning method considers the optimization configuration of muPMU, and obtains the best muPMU configuration result. The electrical variation before and after island micro-grid fault is analyzed. The CNN-LSTM model is constructed by combining the advantages of CNN and LSTM. The input and output of the CNN-LSTM model are constructed by combining the analysis of the electrical variation before and after island micro-grid fault. The shortcoming that the traditional CNN model does not contain a recurrent neuron and cannot remember features is effectively solved, the accuracy of fault positioning is greatly improved, and the loss in the training process is reduced.

Description

TECHNICAL FIELD

[0001] The application relates to the field of fault positioning, and particularly relates to a CNN-LSTM-based island micro-grid fault positioning method. BACKGROUND

[0002] With the large-scale grid connection of distributed power sources (DG) and electric vehicles, the structure of a distribution network is changed from a radial structure to a multi-source and multi-terminal network architecture, and the short-circuit fault characteristics caused by the change will be fundamentally changed, which puts forward higher requirements for distribution network fault positioning. The operation and maintenance of a distribution network need an accurate monitoring system state, and a μPMU can provide more accurate and effective data compared with other measurement devices, and has the characteristics of small size and low cost. However, from the economic point of view, it is impossible to configure a μPMU at each node, so how to effectively configure a μPMU has been the research focus at present. Therefore, it is very important to analyze the active distribution network fault positioning based on the optimal configuration of a μPMU. The most widely used method in the distribution network fault positioning is mainly some matrix algorithms and intelligent algorithms. However, in recent years, with the rise of deep learning, the deep learning is used for fault positioning and has a better effect, so the application provides a CNN-LSTM-based island micro-grid fault positioning method. SUMMARY

[0003] The application aims to provide a CNN-LSTM-based island micro-grid fault positioning method, which solves the above technical problems and has a good positioning effect.

[0004] The application can be realized by the following technical scheme.

[0005] A CNN-LSTM-based island micro-grid fault positioning method comprises the following steps.

[0006] S1, an island micro-grid fault positioning framework is constructed.

[0007] The optimal configuration of a μPMU under the condition of meeting the observability is analyzed, the fault characteristics of an island micro-grid, an island micro-grid fault positioning method, and an island micro-grid fault positioning framework is constructed.

[0008] S2, an island micro-grid μPMU optimal configuration model is constructed.

[0009] The island micro-grid μPMU optimal configuration model is constructed by taking the least number of μPMU configurations as an objective function and taking the observability of an active distribution network as a constraint condition.

[0010] S3, the island micro-grid μPMU optimal configuration model is solved.

[0011] The discrete particle swarm algorithm is used to solve the model.

[0012] S4. Analysis of electrical changes before and after faults in island microgrids;

[0013] Analyze the changes in electrical quantities caused by topological changes before and after a fault in an island microgrid.

[0014] S5 and CNN-LSTM model construction;

[0015] By combining the advantages of CNN and LSTM, a CNN-LSTM model is constructed.

[0016] S6, CNN-LSTM input-output construction;

[0017] The characteristics of the faults are analyzed, and the original data is transformed into a fault set that is more suitable for the algorithm.

[0018] Furthermore, the fault location framework includes a μPMU optimization configuration layer, a fault feature extraction layer, and a fault location layer.

[0019] Furthermore, the μPMU optimization configuration layer sends the optimal node configuration result to the fault feature extraction layer. Based on the optimal nodes provided by the μPMU optimization configuration layer, the fault feature extraction layer obtains the node electrical quantities at the time of the fault. Then, it analyzes the fault characteristics, transforming the raw data into a fault set more suitable for the algorithm. The fault localization layer first combines the advantages of CNN and LSTM to construct a CNN-LSTM model. Then, it uses the fault features sent by the fault feature extraction layer as input to the CNN-LSTM model and the fault point number as output to begin training the CNN-LSTM model.

[0020] Furthermore, in step S2, the optimal configuration model for the island microgrid μPMU is constructed as follows:

[0021] Its model is shown below:

[0022]

[0023]

[0024] In the formula: N represents the number of nodes in the active power distribution network. Since the substation bus outlet is considered separately, i is accumulated starting from node 2. P i_P This refers to whether a μPMU device is configured at the i-th node in the network, P i_P =1 indicates that μPMU is configured at the node, P i_P =0 indicates that the node is not configured with a μPMU; in addition, for safety reasons, a μPMU device needs to be installed at the substation bus outlet of the actual power distribution network, therefore P = 0. i_P =1 is also used as a constraint condition for this optimization model.

[0025] Furthermore, in step S3, the optimal configuration model of the island microgrid μPMU is solved:

[0026] The above model belongs to the "0-1" integer programming problem with strong nonlinearity and nonconvex constraints. Traditional mathematical programming methods are difficult to solve it, so the discrete particle swarm optimization algorithm is used to solve it.

[0027] The velocity and position are updated in the discrete particle swarm optimization algorithm as follows:

[0028] v i (t+1)=v i (t)+c1r1[p i,best (t)-X i (t)]+c2r2[p g,best (t)-X i (t)]

[0029] In the formula: v i (t+1) and v i (t) represents the velocity of the particle at time t and time t+1, respectively. i (t) represents the position of the particle at time t. c1 and c2 are learning factors. r1 and r2 are random numbers uniformly distributed in [0,1]. i,best (t) represents the optimal position of an individual, p g,best (t) represents the optimal position of the group.

[0030] Although the particle's velocity is updated in the same way, its position is updated as follows:

[0031] The sigmoid function is used to map the speed of the example to the range of 0-1:

[0032]

[0033] In the formula: s(v i,j ) represents the probability that the particle's trajectory is currently 1, v i,j Indicates particle velocity.

[0034] The position of the particle can then be represented as:

[0035]

[0036] In the formula: rand() represents a random number uniformly distributed in (0, 1). i,j This represents the position of the particle. s(v) i,j ) represents the probability that the particle trajectory is currently 1.

[0037] Furthermore, the electrical changes before and after the fault in the island microgrid are analyzed in step S4:

[0038] Before the fault occurred, the voltage and current of the system nodes were respectively The system node admittance matrix is ​​Y 0 ∈C n×n We can obtain:

[0039] Y 0 =I 0 U 0

[0040] The main types of faults in active distribution networks include single-phase ground faults, two-phase ground faults, two-phase faults, and three-phase short circuits. Regardless of the type of fault, each will add a fault point to the distribution network topology. Assuming the fault occurs between nodes i and j, and the added fault point is F, then the node admittance matrix of the distribution network is Y. F ∈C (n+1)×(n+1) Its expansion is shown below:

[0041]

[0042] Where Y′ is the admittance matrix of n nodes in the original topology during the fault period before the fault point is added, y f1 It is the mutual admittance between the fault point F and other topology nodes; y f2 It is the self-admittance of the fault point F.

[0043] Suppose that after the fault occurs, the voltage and current matrices of each node are U′∈C n×1 ,I′∈C n×1 Then it satisfies:

[0044]

[0045] We can obtain:

[0046] I′=Y′U′+y f1 U f

[0047] Let Y′=Y 0 -Y″, where Y″ is represented as follows:

[0048]

[0049] After simplification, we get the following formula:

[0050] I′=(Y 0 -Y″)U′+y f1 U f =Y 0 U′-ΔI″

[0051] Where ΔI″=Y″U′-y f1 U f This indicates an unbalanced current.

[0052] The changes in voltage and current before and after the fault are as follows:

[0053]

[0054] Furthermore, in step S5, the CNN-LSTM model is constructed as follows:

[0055] A CNN network is used to extract local fault features, and then an LSTM network is used to extract long-range features from these local features. Finally, fully connected layers and softmax layers are used to classify the faulty lines.

[0056] Furthermore, in step S6, the CNN-LSTM input and output are constructed as follows:

[0057] Combine the following equations:

[0058]

[0059] We can obtain:

[0060] Y 0 ΔU=ΔI″+ΔI

[0061] Therefore, the eigenvector is defined as:

[0062] f = Y 0 ΔU

[0063] By using the fault features as input to the CNN-LSTM and the fault location as output, the input and output of the CNN-LSTM can be constructed.

[0064] The beneficial effects of this invention are:

[0065] 1. The fault location method for island microgrids of the present invention fully considers the optimized configuration of μPMU, provides accurate and effective data for fault location, and also meets the economic requirements, greatly saving costs.

[0066] 2. The fault location method for island microgrids of the present invention analyzes the electrical quantities before and after the fault, determines the fault characteristics, and transforms the original data into a fault set that is more suitable for the algorithm, which greatly improves the speed and accuracy of model training, while reducing losses.

[0067] 3. The fault location method for island microgrids in this invention uses a CNN-LSTM model to learn fault information, and combines the advantages of CNN and LSTM networks, which greatly improves the fault location effect. Attached Figure Description

[0068] The invention will now be further described with reference to the accompanying drawings.

[0069] Figure 1 This invention provides a fault location framework for island microgrids.

[0070] Figure 2 This is a diagram showing the optimized configuration result of the μPMU in this invention.

[0071] Figure 3 This is a comparison chart of node voltage amplitudes under the optimal and random configurations of the μPMU in this invention.

[0072] Figure 4 This is a comparison diagram of node voltage phase angles for state estimation under optimal and random configurations of the μPMU in this invention.

[0073] Figure 5 This is a comparison chart of the absolute error of node voltage amplitude in state estimation under optimal and random configurations of the μPMU of this invention.

[0074] Figure 6 This is a comparison chart of the absolute error of the node voltage phase angle for state estimation under the optimal and random configurations of the μPMU in this invention.

[0075] Figure 7 This is a training diagram of the CNN-LSTM model with Accuracy Iteration of this invention.

[0076] Figure 8 This is a training diagram of the CNN-LSTM model with LOSS iteration in this invention. Detailed Implementation

[0077] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0078] like Figure 1 As shown, a fault location method for island microgrids based on CNN-LSTM includes the following steps:

[0079] (1) Construction of fault location framework for island microgrids;

[0080] The optimal configuration of μPMU under observability conditions is analyzed, the fault characteristics of island microgrids are analyzed, the fault location method of island microgrids is analyzed, and a fault location framework for island microgrids is constructed.

[0081] (2) Construction of the μPMU optimization configuration model for island microgrids;

[0082] With the objective function of minimizing the number of μPMU configurations and the constraint of satisfying the observability of the active power distribution network, an optimal configuration model for μPMUs in an island microgrid is constructed.

[0083] (3) Solving the optimal configuration model of the island microgrid μPMU;

[0084] The discrete particle swarm optimization algorithm is used to solve it.

[0085] (4) Analysis of electrical changes before and after faults in island microgrids;

[0086] Analyze the changes in electrical quantities caused by topological changes before and after a fault in an island microgrid.

[0087] (5) Construction of CNN-LSTM model;

[0088] By combining the advantages of CNN and LSTM, a CNN-LSTM model is constructed.

[0089] (6) Constructing the input and output of CNN-LSTM;

[0090] The characteristics of the faults are analyzed, and the original data is transformed into a fault set that is more suitable for the algorithm.

[0091] In step (1), the fault location framework for island microgrids is constructed. Accurate and effective data is crucial for fault location in island microgrids, and cost considerations are also essential when acquiring data. Furthermore, when collecting a large amount of data, selecting suitable feature data for the algorithm to make the training process faster and more accurate is also a problem. Therefore, a μPMU optimization configuration layer, a fault feature extraction layer, and a fault location layer need to be established.

[0092] The μPMU optimization configuration layer sends the optimal node configuration result to the fault feature extraction layer. Based on the optimal nodes provided by the μPMU optimization configuration layer, the fault feature extraction layer obtains the node electrical quantities at the time of the fault. Next, it analyzes the fault characteristics, transforming the raw data into a fault set more suitable for the algorithm. The fault localization layer first combines the advantages of CNN and LSTM to construct a CNN-LSTM model. Then, it uses the fault features sent by the fault feature extraction layer as input to the CNN-LSTM model and the fault point number as output to begin training the CNN-LSTM model.

[0093] In step (2), the construction of the μPMU optimal configuration model for the island microgrid is crucial. Compared with other measurement devices, μPMUs can provide more accurate data, which can provide precise and effective data for fault location. However, they are expensive and cannot be configured globally, so they need to be optimized under the conditions of meeting economic and observability requirements. Therefore, the construction of the μPMU optimal configuration model for the island microgrid is very important.

[0094] Its model is shown below:

[0095]

[0096]

[0097] In the formula: N represents the number of nodes in the active power distribution network. Since the substation bus outlet is considered separately, i is accumulated starting from node 2. P i_P This refers to whether a μPMU device is configured at the i-th node in the network, P i_P =1 indicates that μPMU is configured at the node, P i_P =0 indicates that the node is not configured with a μPMU; in addition, for safety reasons, a μPMU device needs to be installed at the substation bus outlet of the actual power distribution network, therefore P = 0. i_P =1 is also used as a constraint condition for this optimization model.

[0098] In step (3), the μPMU optimization configuration model of the island microgrid is solved. The μPMU optimization configuration model belongs to the "0-1" integer programming problem with strong nonlinearity and non-convex constraints. Traditional mathematical programming methods are difficult to solve it, so the discrete particle swarm algorithm is used to solve it.

[0099] The velocity and position are updated in the discrete particle swarm optimization algorithm as follows:

[0100] v i (t+1)=v i (t)+c1r1[p i,best (t)-X i (t)]+c2r2[p g,best (t)-X i (t)]

[0101] In the formula: v i (t+1) and v i (t) represents the velocity of the particle at time t and time t+1, respectively. i (t) represents the position of the particle at time t. c1 and c2 are learning factors. r1 and r2 are random numbers uniformly distributed in [0,1]. i,best (t) represents the optimal position of an individual, p g,best (t) represents the optimal position of the group.

[0102] Although the particle's velocity is updated in the same way, its position is updated as follows:

[0103] The sigmoid function is used to map the speed of the example to the range of 0-1:

[0104]

[0105] In the formula: s(v i,j ) represents the probability that the particle's trajectory is currently 1, v i,j Indicates particle velocity.

[0106] The position of the particle can then be represented as:

[0107]

[0108] In the formula: rand() represents a random number uniformly distributed in (0, 1). i,j This represents the position of the particle. s(v) i,j ) represents the probability that the particle trajectory is currently 1.

[0109] In step (4), we analyze the electrical changes before and after the fault in the island microgrid. Before and after the fault, the electrical quantities and topology of the island microgrid will change significantly. Analyzing these changes can help us find a more suitable feature dataset for the algorithm and effectively improve the training effect. Therefore, we need to analyze the electrical changes before and after the fault in the island microgrid.

[0110] Before the fault occurred, the voltage and current of the system nodes were respectively The system node admittance matrix is ​​Y 0 ∈C n×n We can obtain:

[0111] Y 0 =I 0 U 0

[0112] The main types of faults in active distribution networks include single-phase ground faults, two-phase ground faults, two-phase faults, and three-phase short circuits. Regardless of the type of fault, each will add a fault point to the distribution network topology. Assuming the fault occurs between nodes i and j, and the added fault point is F, then the node admittance matrix of the distribution network is Y. F ∈C (n+1)×(n+1) Its expansion is shown below:

[0113]

[0114] Where Y′ is the admittance matrix of n nodes in the original topology during the fault period before the fault point is added, y f1 It is the mutual admittance between the fault point F and other topology nodes; y f2 It is the self-admittance of the fault point F.

[0115] Suppose that after the fault occurs, the voltage and current matrices of each node are U′∈C n×1 ,I′∈C n×1 Then it satisfies:

[0116]

[0117] We can obtain:

[0118] I′=Y′U′+y f1 U f

[0119] Let Y′=Y 0 -Y″, where Y″ is represented as follows:

[0120]

[0121] After simplification, we get:

[0122] I′=(Y 0 -Y″)U′+y f1 U f =Y 0 U′-ΔI″

[0123] Where ΔI″=Y″U′-y f1 U f This indicates an unbalanced current.

[0124] The changes in voltage and current before and after the fault are as follows:

[0125]

[0126] Regarding step (5), in constructing the CNN-LSTM model, CNN is a relatively mature algorithm that can learn corresponding fault features from a large amount of data, but it does not contain recurrent neurons, which means it does not have memory features. LSTM can solve this problem well, so we need to construct the CNN-LSTM model.

[0127] A CNN network is used to extract local fault features, and then an LSTM network is used to extract long-range features from these local features. Finally, fully connected layers and softmax layers are used to classify the faulty lines.

[0128] Regarding step (6), which involves constructing the input and output of the CNN-LSTM, once we have built the CNN-LSTM model, we need to select a suitable input and output to make the training process more effective.

[0129] Its input and output are set as follows:

[0130] Combine the following equations:

[0131]

[0132] We can obtain:

[0133] Y 0 ΔU=ΔI″+ΔI

[0134] Therefore, the eigenvector is defined as:

[0135] f = Y 0 ΔU

[0136] By using the fault features as input to the CNN-LSTM and the fault location as output, the input and output of the CNN-LSTM can be constructed.

[0137] Figure 2 This is the result of μPMU optimization configuration. The discrete particle swarm optimization algorithm was used in Matlab for μPMU optimization configuration. The configuration result is: μPMUs are configured at nodes 1, 4, 19, 25, 30, and 32.

[0138] Figure 3 , Figure 4 , Figure 5 , Figure 6 These are comparison charts showing the node voltage magnitude, node voltage phase angle, absolute error of node voltage magnitude, and absolute error of node voltage phase angle for state estimation under optimal and random configurations of the μPMU.

[0139] Since there are 6 nodes with optimal configuration of μPMU proposed in this invention under the condition of satisfying the observability of the distribution network, we also select 6 nodes when randomly configuring μPMU. Considering the uniformity of distribution, we choose to configure μPMU at nodes 1, 6, 12, 18, 24, and 30.

[0140] The information in the figure clearly shows that the voltage amplitude and voltage phase angle results of the optimally configured μPMU have smaller errors than those of the randomly configured μPMU, which proves the effectiveness of the proposed method. Furthermore, the absolute errors of the voltage amplitude and voltage phase angle of the optimally configured μPMU and the randomly configured μPMU are compared. The error results demonstrate the accuracy of the proposed optimally configured μPMU method.

[0141] Table 1 shows the location accuracy of the fault location method proposed in this invention under different faults.

[0142] Table 1

[0143]

[0144] The results show that the accuracy rate of fault location exceeds 90% for most faults, demonstrating the effectiveness of the method. Furthermore, it can be seen that this method achieves the highest accuracy rate for locating two-phase ground faults.

[0145] Figure 7 , Figure 8This is a training graph of the CNN-LSTM model. As can be seen from the graph, the initial loss is large, but it gradually approaches 0 as the number of iterations increases. Meanwhile, the accuracy increases with the number of iterations, eventually reaching a relatively high value. This verifies the effectiveness of the method proposed in this invention.

[0146] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0147] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention.

Claims

1. A CNN-LSTM-based island microgrid fault location method, characterized in that, The method comprises the following steps: S1, constructing a fault positioning framework of an island micro-grid; Considering the optimal configuration of the μPMU under the condition of meeting the observability, analyzing the fault characteristics of the island micro-grid, determining the fault positioning method of the island micro-grid, and constructing a fault positioning framework of the island micro-grid; S2, constructing an optimal configuration model of the μPMU of the island micro-grid; An optimal configuration model of the μPMU of the island micro-grid is constructed, with the minimum number of the configured μPMU as an objective function and the observability of the active distribution network as a constraint condition; S3, solving the optimal configuration model of the μPMU of the island micro-grid; The discrete particle swarm algorithm is used to solve the model; S4, analyzing the electrical variation before and after the fault of the island micro-grid; The electrical variation caused by the topological change before and after the fault of the island micro-grid is analyzed; S5, constructing a CNN-LSTM model; A CNN-LSTM model is constructed by combining the advantages of the CNN and the LSTM; S6, constructing the input and output of the CNN-LSTM; The original data is transformed into a fault set more suitable for the algorithm by analyzing the characteristics of the fault; The fault positioning framework comprises a μPMU optimal configuration layer, a fault feature extraction layer and a fault positioning layer; The μPMU optimal configuration layer sends the optimal node configuration result to the fault feature extraction layer; The fault feature extraction layer obtains the electrical quantity of the node when the fault occurs according to the optimal node provided by the μPMU optimal configuration layer, analyzes the characteristics of the fault, and transforms the original data into a fault set more suitable for the algorithm; the fault positioning layer first constructs a CNN-LSTM model by combining the advantages of the CNN and the LSTM, then takes the fault feature sent by the fault feature extraction layer as the input of the CNN-LSTM model, takes the fault point number as the output, and starts training the CNN-LSTM model; In step S2, the optimal configuration model of the μPMU of the island micro-grid is constructed as follows: The objective function of the optimal model of the μPMU is the minimum number of the configured μPMU, and the constraint condition is the observability of the active distribution network; the model is shown as follows: where N represents the number of nodes in the active power distribution network, i is accumulated from node 2 since the substation bus outlet is considered separately, denotes whether a μPMU device is configured at the i th node in the network, denotes that a μPMU is configured at the node, denotes that the node is not configured with a μPMU; in addition, for safety considerations, a μPMU device needs to be installed at the substation bus outlet of the actual power distribution network, so that is also taken as a constraint condition of the optimization model; In step S4, the analysis of the electrical variation before and after the fault of the island micro-grid is as follows: Before the fault occurs, the voltage and current of the system node are , , the admittance matrix of the system node is , and it can be obtained that The active distribution network fault types mainly include single-phase grounding fault, two-phase grounding short circuit, two-phase fault and three-phase short circuit; no matter which fault is above, the distribution network topology structure will increase a fault point; assuming that the fault occurs between the i and j nodes, the increased fault point is F, and the node admittance matrix of the distribution network is The expansion formula is as follows: where Y' is the admittance matrix of the n nodes in the original topology before the introduction of the fault point, during the fault, is the mutual admittance between the fault point F and the other nodes of the topology; is the self-admittance of the fault point F. After the fault occurs, the voltage and current matrix of each node is , , then it satisfies: After simplification, the following formula is obtained: Let where is represented as follows: The variation of the voltage and current before and after the fault is respectively as follows: wherein represents an unbalanced current; In step S6, the input and output of the CNN-LSTM are constructed as follows: The following formula is solved: The following formula is obtained: Therefore, the feature vector is defined as: The fault feature is taken as the input of the CNN-LSTM, and the fault position is taken as the output, so that the input and output of the CNN-LSTM are constructed. In step S3, the optimal configuration model of the μPMU of the island micro-grid is solved as follows:

2. The CNN-LSTM-based fault location method for an island microgrid according to claim 1, characterized in that, The optimal configuration model of the μPMU belongs to a "0-1" integer programming problem with strong nonlinearity and non-convex constraints, and it is difficult to solve by using traditional mathematical programming methods; the discrete particle swarm algorithm is used to solve the model; The updating mode of the speed and position of the discrete particle swarm algorithm is as follows: Although the updating mode of the speed of the particle is unchanged, the updating mode of the position of the particle is as follows: wherein: and denote the velocity of the particle at time t and t+1, respectively, denotes the position of the particle at time t, , is a learning factor, and are random numbers uniformly distributed in [0,1], denotes the individual best position, denotes the swarm best position; The speed of the particle is mapped to 0-1 by using the sigmoid function: The position of the particle can be represented as: wherein: denotes the probability that the particle trajectory is currently 1, denotes the particle velocity; In step S5, the CNN-LSTM model is constructed as follows: where: represents a random number uniformly distributed in (0, 1), represents the position of a particle, represents the probability that the particle trajectory is currently 1.

3. The CNN-LSTM-based fault location method for island microgrid according to claim 1, characterized in that, ​ First, the local fault features are extracted by CNN network, and then the long-distance features of these local features are extracted by LSTM network. Finally, the fault line is classified by full connection layer and softmax layer.

Images