EMD-PCA-LSTM-based power distribution network reliability evaluation method and system

Abstract

The invention relates to a power distribution network reliability evaluation method and system based on EMD-PCA-LSTM, belongs to the technical field of electric digital data processing, and aims at integrating distributed energy into a power distribution network at an unprecedented speed under the driving of global energy transformation and carbon neutralization targets and simultaneously bringing a certain influence on the reliable operation of the power distribution network. A traditional statistical reasoning model is difficult to process the non-stable and multi-dimensional influence of power grid operation data and environmental data, so that the coverage range and prediction precision of power distribution network reliability evaluation have significant limitations. Aiming at the problem, the method comprises the following steps: firstly, decomposing original data into a multi-scale intrinsic mode function (IMF) by adopting empirical mode decomposition (EMD); secondly, dimension reduction is carried out on IMF components through principal component analysis (PCA), redundancy is eliminated, and core features influencing reliability are extracted; and finally, constructing a long short-term memory (LSTM) network model, learning a complex time sequence dependency relationship between the core characteristics and the reliability indexes by using a gating mechanism, and realizing accurate prediction of the reliability of the power distribution network.

Description

Technical Field

[0001] This invention belongs to the field of electrical digital data processing technology, specifically relating to a method and system for reliability assessment of distribution networks based on EMD-PCA-LSTM. Background Technology

[0002] Driven by the global energy transition and carbon neutrality goals, distributed energy resources are being integrated into distribution networks at an unprecedented pace, while also impacting the reliable operation of these networks. Traditional statistical inference models struggle to handle the non-stationary and multi-dimensional effects of grid operation and environmental data, resulting in significant limitations in the coverage and prediction accuracy of distribution network reliability assessments.

[0003] Therefore, at this stage, it is necessary to design a distribution network reliability assessment method and system based on EMD-PCA-LSTM to solve the above problems. Summary of the Invention

[0004] The purpose of this invention is to provide a distribution network reliability assessment method and system based on EMD-PCA-LSTM to solve the technical problems existing in the prior art. Traditional statistical inference models are difficult to handle the non-stationary and multi-dimensional effects of power grid operation data and environmental data, which makes the coverage and prediction accuracy of distribution network reliability assessment significantly limited.

[0005] To achieve the above objectives, the technical solution of the present invention is as follows:

[0006] A distribution network reliability assessment method based on EMD-PCA-LSTM includes the following steps:

[0007] S1. First, Empirical Mode Decomposition (EMD) is used to decompose the original data into multi-scale intrinsic mode functions (IMFs).

[0008] S2. Secondly, principal component analysis (PCA) is used to reduce the dimensionality of the multi-scale intrinsic mode function (IMF) components, eliminate redundancy, and extract the core features that affect reliability.

[0009] S3. Finally, a Long Short-Term Memory (LSTM) network model is constructed, and a gating mechanism is used to learn the complex temporal dependencies between core features and reliability indicators, so as to achieve accurate prediction of the reliability of the distribution network.

[0010] Furthermore, in step S1, the data needs to be preprocessed, as follows:

[0011] (1) Handling missing values

[0012] 1) Short-term missing data filling, using linear interpolation:

[0013]

[0014] In the formula: x t The value filled in at time t; k is the number of missing points;

[0015] 2) Long-term missing data filling:

[0016] Periodic data are analyzed using the historical average method for the same period, as shown in equation (2); non-periodic data are analyzed using the moving average method, that is, using the average value of the variable over a recent period, as shown in equation (3):

[0017]

[0018] In the formula: T is the period of 24 hours; N is the number of days of historical data; m is the size of the sliding window;

[0019] (2) Outlier handling

[0020] 1) Physical threshold filtering: Set reasonable ranges for each variable based on physical laws; values ​​outside the range are considered outliers and corrected to boundary values.

[0021] 2) Statistical threshold filtering:

[0022] For data that follows a Gaussian distribution, outlier correction is performed using the mean and standard deviation. The mean μ and standard deviation σ of each variable are calculated, and values ​​exceeding [μ-3σ, μ+3σ] are considered outliers and corrected to the mean μ; as shown in equation (4).

[0023]

[0024] 3) Mutation point correction:

[0025] For data points with excessively large changes between adjacent time points, a difference threshold method is used for correction. The absolute difference between adjacent time points is calculated, and the 99th percentile of this difference is used as the threshold. If the difference at a point exceeds the threshold, the value at that point is replaced with the value x from the previous time point. t-1 The formula is:

[0026] |Δx t |=|x t -x t-1 |>Q 0.99 (Δx) (5).

[0027] Furthermore, in step S1, the Empirical Mode Decomposition (EMD) is as follows:

[0028] Find all the maximum and minimum points of the original data sequence x(t), and use cubic spline interpolation function to fit the upper and lower envelopes of the original data. The mean of the upper and lower envelopes is denoted as m(t). Subtract the envelope mean from the original data sequence to obtain a new data sequence h(t) = x(t) - m(t).

[0029] If h(t) satisfies the conditions of the IMF, then h(t) is treated as an IMF component I. i (t) is extracted: Let I i (t) = h(t); if h(t) does not satisfy the IMF condition, then h(t) is regarded as the new original data, and step 1) is repeated to continue the screening iteration;

[0030] Repeat the sieving process n times until the remaining term r is reached. n (t) is a monotonic function or a constant;

[0031] After n sieving processes, x(t) is decomposed into n IMF components extracted in descending order of frequency. and a residual term r n (t), as in equation (6);

[0032]

[0033] Furthermore, in step S2, the principal component analysis is performed as follows:

[0034] Construct a standardized matrix of the original data;

[0035] The covariance matrix is ​​obtained by calculating the covariance of the standardized data; the eigenvalues ​​λ1, λ2, ..., λ of the covariance matrix are then calculated. m and eigenvectors μ1,μ2,...,μ m ;

[0036] Calculate each eigenvalue λ i The variance contribution rate and the cumulative variance contribution rate of the first k principal components are shown in equations (7) and (8);

[0037]

[0038] By selecting the eigenvalues ​​of the principal components and their corresponding eigenvectors, a new data matrix with reduced dimensionality is obtained, thus completing the dimensionality reduction process of the data.

[0039] Furthermore, in step S3, the Long Short-Term Memory (LSTM) network model is specifically as follows:

[0040] Forget Gate: The forget gate is the output h of the layer above. t-1 and the sequence data x to be input to this layer t As input, it passes through a sigmoid activation function σ, and the output is f. t ;f t The output value takes the range [0,1], representing the probability that the cell state of the previous layer is forgotten, where 1 means "completely retained" and 0 means "completely discarded"; its expression is:

[0041] ft =σ(W f ·[h t-1 ,x t ]+b f (9)

[0042] In the formula: W f b is the weight matrix of the forget gate; f This is the bias vector for the forget gate;

[0043] Input gate: The input gate consists of two parts. The first part uses the sigmoid activation function σ, and the output is i. t The first part takes values ​​in the range [0,1]; the second part uses the tanh activation function, and the output is... Its expression is:

[0044] i t =σ(W i ·[h t-1 ,x t ]+b i (10)

[0045]

[0046] In the formula: W i b is the weight matrix of the input gate, used to control which new information should be updated to the cell state; i W is the bias vector of the input gate; C b is the weight matrix for candidate cell states, used to generate new candidate information; C is the bias vector for the candidate cell state; C represents the candidate cell state vector, with values ​​in the interval [-1, 1]. t-1 This represents the cell state at the previous time step, carrying long-term memory information; C t The updated current cell state combines the results of the forget gate and the input gate.

[0047] Output gate: The output gate controls how many cell states in this layer are filtered; first, the sigmoid activation function is used to obtain an o value in the range [0,1]. t Next, the cell state C t After processing with the tanh activation function, and then with o t Multiplication is the output h of this layer. t Its expression is:

[0048] o t =σ(W o ·[h t-1 ,x t ]+b o (13)

[0049] h t =o t *tanh(C t (14)

[0050] In the formula: W o b is the weight matrix of the output gate, used to control how much information about the cell state is output to the hidden state; o The bias vector of the output gate; o t h is the output vector of the output gate, with values ​​in the interval [0,1]. t The hidden state at the current time step is the output of the LSTM, which is passed to the next time step or used for prediction.

[0051] The prediction result is as follows:

[0052] y t =σ(W y ·h t +b y (15)

[0053] In the formula: W y b is the weight matrix of the output layer; y y is the bias vector of the output layer; t This is the prediction output for the current time step.

[0054] A distribution network reliability assessment method based on EMD-PCA-LSTM includes the following steps:

[0055] Step 1: Complete and clean the collected power grid operation data and environmental data that may be related to the reliability of the distribution network to construct the original dataset;

[0056] Step 2: Decompose the original time series data into multiple intrinsic mode functions (IMFs) and a residual term through EMD decomposition;

[0057] Step 3: Use PCA to reduce the dimensionality and extract features from the multiple IMF components obtained in Step 2, and retain the top k principal components with a cumulative variance contribution rate greater than 85% as the core features affecting the reliability of the distribution network.

[0058] Step 4: Using the dimensionality reduction data obtained in Step 3, construct the final supervised learning dataset for LSTM, and divide it into training set, validation set, and test set;

[0059] Step 5: Initialize the LSTM network, input the training set data, update the weights through backpropagation, and adjust the hyperparameters using the validation set;

[0060] Step 6: Reliability assessment and prediction. Use the test set to evaluate the model performance and output reliability assessment metrics, including average outage duration (SAIDI) and average outage frequency (SAIFI).

[0061] A distribution network reliability assessment system based on EMD-PCA-LSTM is provided, wherein the distribution network reliability assessment system based on EMD-PCA-LSTM is applied to the distribution network reliability assessment method based on EMD-PCA-LSTM described above.

[0062] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0063] First, Empirical Mode Decomposition (EMD) is used to decompose the original data into multi-scale intrinsic mode functions (IMFs). Second, Principal Component Analysis (PCA) is used to reduce the dimensionality of the IMF components, eliminate redundancy, and extract the core features affecting reliability. Finally, a Long Short-Term Memory (LSTM) model is constructed, and a gating mechanism is used to learn the complex temporal dependencies between core features and reliability indicators, achieving accurate prediction of distribution network reliability. Principal Component Analysis is an unsupervised linear dimensionality reduction and feature extraction technique. Its purpose is to reduce the dimensionality of the dataset while retaining the most important features and removing noise and unimportant features, thereby improving data processing speed and accuracy. The core idea of ​​LSTM is to introduce a carefully designed memory unit and three gating mechanisms to precisely control the flow of information. This structure allows it to selectively retain, update, and transmit information, thus overcoming the long-term dependency problem. Attached Figure Description

[0064] Figure 1 This is a schematic diagram illustrating the basic steps of an embodiment of this solution.

[0065] Figure 2 This is a schematic diagram illustrating the solution steps of an embodiment of this solution. Detailed Implementation

[0066] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only for explaining the invention and are not intended to limit the invention; that is, the described embodiments are merely some embodiments of the invention, and not all embodiments.

[0067] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention. It should be noted that relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations.

[0068] Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0069] Example:

[0070] Driven by the global energy transition and carbon neutrality goals, distributed energy resources are being integrated into distribution networks at an unprecedented pace, while also impacting the reliable operation of these networks. Traditional statistical inference models struggle to handle the non-stationary and multi-dimensional effects of grid operation and environmental data, resulting in significant limitations in the coverage and prediction accuracy of distribution network reliability assessments.

[0071] To address this issue, a distribution network reliability assessment method based on EMD-PCA-LSTM is proposed, such as... Figure 1 As shown, firstly, Empirical Mode Decomposition (EMD) is used to decompose the original data into multi-scale intrinsic mode functions (IMFs); secondly, Principal Component Analysis (PCA) is used to reduce the dimensionality of the IMF components, eliminate redundancy, and extract the core features that affect reliability; finally, a Long Short-Term Memory (LSTM) network model is constructed, and the complex temporal dependency relationship between the core features and reliability indicators is learned using a gating mechanism to achieve accurate prediction of the reliability of the distribution network.

[0072] 1. Data Preprocessing

[0073] (1) Handling missing values

[0074] 1) Short-term missing data imputation (continuous missing data ≤ 3 points), using linear interpolation:

[0075]

[0076] In the formula: xt t represents the fill value at time t; k represents the number of missing points.

[0077] 2) Long-term missing data imputation (continuous missing data > 3 points):

[0078] Periodic data are analyzed using the historical average method for the same period, as shown in equation (2); non-periodic data are analyzed using the moving average method, that is, using the average value of the variable over a recent period, as shown in equation (3):

[0079]

[0080] In the formula: T is the period of 24 hours; N is the number of days of historical data; m is the sliding window size, which is usually taken as 5.

[0081] (2) Outlier handling

[0082] 1) Physical threshold filtering: Set reasonable ranges for each variable based on physical laws. Values ​​outside the range are considered outliers and corrected to boundary values.

[0083] 2) Statistical threshold filtering:

[0084] For data that follows a Gaussian distribution, outlier correction is performed using the mean and standard deviation. The mean μ and standard deviation σ of each variable are calculated, and values ​​exceeding [μ-3σ, μ+3σ] are considered outliers and corrected to the mean μ.

[0085] As in equation (4)

[0086]

[0087] 3) Mutation point correction:

[0088] For data points where the changes are too large between adjacent time points, a difference threshold method is used for correction. The absolute difference between adjacent time points is calculated, and the 99th percentile of this difference is used as the threshold. If the difference at a point exceeds the threshold, the value at that point is replaced with the value x from the previous time point. t-1 The formula is:

[0089] |Δx t |=|x t -x t-1 |>Q 0.99 (Δx) (5)

[0090] 2 Empirical Mode Decomposition

[0091] The basic principle of EMD is to decompose a complex signal into a finite number of intrinsic mode functions (IMFs). Each IMF component contains local characteristic signals of the original signal at different time scales. Power grid operation data and environmental data are inherently non-stationary, as shown in Table 1. Through EMD decomposition, characteristic fluctuation patterns at different time scales are obtained: high-frequency IMFs reflect short-term random disturbances; mid-frequency IMFs reflect periodic fluctuations; and low-frequency IMFs characterize long-term trends.

[0092] Table 1 Feature Data Table

[0093]

[0094] The specific decomposition process is as follows:

[0095] 5) Find all the maximum and minimum points of the original data sequence x(t), and use cubic spline interpolation function to fit the upper and lower envelopes of the original data. The mean of the upper and lower envelopes is denoted as m(t). Subtract the envelope mean from the original data sequence to obtain a new data sequence h(t) = x(t) - m(t);

[0096] 6) If h(t) satisfies the conditions of the IMF, then it is treated as an IMF component I. i (t) is extracted: Let I i (t) = h(t); if h(t) does not satisfy the IMF condition, then h(t) is regarded as the new original data, and step 1) is repeated to continue the screening iteration;

[0097] 7) Repeat the sieving process n times until the residual item r is reached. n (t) is a monotonic function or a constant;

[0098] 8) After n sieving steps, x(t) is decomposed into n IMF components extracted in descending order of frequency. and a residual term r n (t), as in equation (6).

[0099]

[0100] 3. Principal Component Analysis

[0101] The original data after EMD decomposition is transformed into multiple sets of IMF components, significantly increasing the number of feature sequences. However, the increased dimensionality of the input variables leads to a decrease in computational efficiency and poses a risk of overfitting. Therefore, PCA is introduced to reduce the dimensionality of the input variables.

[0102] Principal Component Analysis (PCA) is an unsupervised linear dimensionality reduction and feature extraction technique. Its purpose is to reduce the dimensionality of a dataset while retaining the most important features and removing noise and unimportant features, thereby improving data processing speed and accuracy. The algorithm's computation process is as follows.

[0103] 4) Construct a standardized matrix of the original data;

[0104] 5) Obtain the covariance matrix by calculating the covariance of the standardized data; calculate the eigenvalues ​​λ1, λ2, ..., λ of the covariance matrix. m and eigenvectors μ1,μ2,...,μ m ;

[0105] 6) Calculate each eigenvalue λ i The variance contribution rate of the first k principal components and the cumulative variance contribution rate of the first k principal components are shown in Equations (7) and (8). Usually, the first k principal components are selected so that the cumulative variance contribution rate is between 80% and 95%.

[0106]

[0107] 7) By selecting the eigenvalues ​​of the principal components and their corresponding eigenvectors, a new data matrix after dimensionality reduction is obtained, thus completing the dimensionality reduction process of the data.

[0108] 4. Long Short-Term Memory Neural Network

[0109] LSTM is a special variant of Recurrent Neural Networks (RNNs) specifically designed to address the vanishing and exploding gradient problems encountered by standard RNNs when processing long sequences of data. The core idea of ​​LSTM is to introduce a carefully designed memory unit and three gating mechanisms to precisely control the flow of information. This structure allows it to selectively retain, update, and pass information, thereby overcoming the long-term dependency problem.

[0110] 4) Forget Gate: The forget gate is the output h of the layer above. t-1 and the sequence data x to be input to this layer t As input, it passes through a sigmoid activation function σ, and the output is f. t f t The output value takes a value in the interval [0,1], representing the probability that the cell state of the previous layer is forgotten; 1 means "completely retained," and 0 means "completely discarded." Its expression is:

[0111] f t =σ(W f ·[h t-1 ,x t ]+b f (9)

[0112] In the formula: W f b is the weight matrix of the forget gate; f This is the bias vector for the forget gate.

[0113] 5) Input gate: The input gate consists of two parts. The first part uses the sigmoid activation function σ, and the output is i. t The first part takes values ​​in the range [0,1]; the second part uses the tanh activation function, and the output is... Its expression is:

[0114] i t =σ(W i ·[h t-1 ,x t ]+b i (10)

[0115]

[0116] In the formula: W i b is the weight matrix of the input gate, used to control which new information should be updated to the cell state; i W is the bias vector of the input gate; C b is the weight matrix for candidate cell states, used to generate new candidate information; C is the bias vector for the candidate cell state; C represents the candidate cell state vector, with values ​​in the interval [-1, 1]. t-1 This represents the cell state at the previous time step, carrying long-term memory information; C t The updated current cell state combines the results of the forget gate and the input gate.

[0117] 6) Output Gate: The output gate controls how many cell states in this layer are filtered. First, a sigmoid activation function is used to obtain an o value that takes values ​​in the range [0,1]. t Next, the cell state C t After processing with the tanh activation function, and then with o t Multiplication is the output h of this layer. t Its expression is:

[0118] o t =σ(W o ·[h t-1 ,x t ]+b o (13)

[0119] h t =o t *tanh(C t (14)

[0120] In the formula: Wo b is the weight matrix of the output gate, used to control how much information about the cell state is output to the hidden state; o The bias vector of the output gate; o t h is the output vector of the output gate, with values ​​in the interval [0,1]. t The hidden state at the current time step is the output of the LSTM, which is passed to the next time step or used for prediction.

[0121] The prediction result is as follows:

[0122] y t =σ(W y ·h t +b y (15)

[0123] In the formula: W y b is the weight matrix of the output layer; y y is the bias vector of the output layer; t This is the prediction output for the current time step.

[0124] 5. Solution process, as follows Figure 2 As shown:

[0125] Step 1: Complete and clean the collected power grid operation data and environmental data that may be related to the reliability of the distribution network to construct the original dataset;

[0126] Step 2: Decompose the original time series data into multiple intrinsic mode functions (IMFs) and a residual term through EMD decomposition;

[0127] Step 3: Use PCA to reduce the dimensionality and extract features from the multiple IMF components obtained in Step 2, and retain the top k principal components with a cumulative variance contribution rate greater than 85% as the core features affecting the reliability of the distribution network.

[0128] Step 4: Using the dimensionality reduction data obtained in Step 3, construct the final supervised learning dataset for LSTM, and divide it into training set, validation set, and test set;

[0129] Step 5: Initialize the LSTM network, input the training set data, update the weights through backpropagation, and adjust the hyperparameters using the validation set;

[0130] Step 6: Reliability assessment and prediction. Use the test set to evaluate the model performance and output reliability assessment metrics such as SAIDI (mean outage duration) and SAIFI (mean outage frequency).

[0131] The above are preferred embodiments of the present invention. Any changes made to the technical solution of the present invention that do not exceed the scope of the technical solution of the present invention shall fall within the protection scope of the present invention.

Claims

1. A distribution network reliability assessment method based on EMD-PCA-LSTM, characterized in that, Includes the following steps: S1. First, Empirical Mode Decomposition (EMD) is used to decompose the original data into multi-scale intrinsic mode functions (IMFs). S2. Secondly, principal component analysis (PCA) is used to reduce the dimensionality of the multi-scale intrinsic mode function (IMF) components, eliminate redundancy, and extract the core features that affect reliability. S3. Finally, a Long Short-Term Memory (LSTM) network model is constructed, and a gating mechanism is used to learn the complex temporal dependencies between core features and reliability indicators, so as to achieve accurate prediction of the reliability of the distribution network.

2. The distribution network reliability assessment method based on EMD-PCA-LSTM according to claim 1, characterized in that, In step S1, the data needs to be preprocessed, as follows: (1) Handling missing values 1) Short-term missing data filling, using linear interpolation: In the formula: x t The value filled in at time t; k is the number of missing points; 2) Long-term missing data filling: Periodic data are analyzed using the historical average method for the same period, as shown in equation (2); non-periodic data are analyzed using the moving average method, that is, using the average value of the variable over a recent period, as shown in equation (3): In the formula: T is the period of 24 hours; N is the number of days of historical data; m is the size of the sliding window; (2) Outlier handling 1) Physical threshold filtering: Set reasonable ranges for each variable based on physical laws; values ​​outside the range are considered outliers and corrected to boundary values. 2) Statistical threshold filtering: For data that follows a Gaussian distribution, outlier correction is performed using the mean and standard deviation. The mean μ and standard deviation σ of each variable are calculated, and values ​​exceeding [μ-3σ, μ+3σ] are considered outliers and corrected to the mean μ; as shown in equation (4). 3) Mutation point correction: For data points with excessively large changes between adjacent time points, a difference threshold method is used for correction. The absolute difference between adjacent time points is calculated, and the 99th percentile of this difference is used as the threshold. If the difference at a point exceeds the threshold, the value at that point is replaced with the value x from the previous time point. t-1 The formula is: |Δx t |=|x t -x t-1 |>Q 0.99 (Δx) (5)。 3. The distribution network reliability assessment method based on EMD-PCA-LSTM according to claim 2, characterized in that, In step S1, the Empirical Mode Decomposition (EMD) is performed as follows: 1) Find all the maximum and minimum points of the original data sequence x(t), and use cubic spline interpolation function to fit the upper and lower envelopes of the original data. The mean of the upper and lower envelopes is denoted as m(t). Subtract the envelope mean from the original data sequence to obtain a new data sequence h(t) = x(t) - m(t); 2) If h(t) satisfies the conditions of the IMF, then h(t) is treated as an IMF component I. i (t) is extracted: Let I i (t) = h(t); if h(t) does not satisfy the IMF condition, then h(t) is regarded as the new original data, and step 1) is repeated to continue the screening iteration; 3) Repeat the sieving process n times until the residual item r is reached. n (t) is a monotonic function or a constant; 4) After n sieving steps, x(t) is decomposed into n IMF components I extracted in descending order of frequency. i (t), (i = 1, 2, ..., n), and a residual term r n (t), as in equation (6); 4. The distribution network reliability assessment method based on EMD-PCA-LSTM according to claim 3, characterized in that, In step S2, the principal component analysis is performed as follows: 1) Construct a standardized matrix of the original data; 2) Obtain the covariance matrix by calculating the covariance of the standardized data; calculate the eigenvalues ​​λ1, λ2, ..., λ of the covariance matrix. m and eigenvectors μ1,μ2,...,μ m ; 3) Calculate each eigenvalue λ i The variance contribution rate and the cumulative variance contribution rate of the first k principal components are shown in equations (7) and (8); By selecting the eigenvalues ​​of the principal components and their corresponding eigenvectors, a new data matrix with reduced dimensionality is obtained, thus completing the dimensionality reduction process of the data.

5. The distribution network reliability assessment method based on EMD-PCA-LSTM according to claim 4, characterized in that, In step S3, the Long Short-Term Memory (LSTM) network model is as follows: 1) Forget Gate: The forget gate is the output h of the layer above. t-1 and the sequence data x to be input to this layer t As input, it passes through a sigmoid activation function σ, and the output is f. t ;f t The output value takes a value in the range [0,1], representing the probability that the cell state of the previous layer is forgotten, where 1 means "completely retained" and 0 means "completely discarded"; its expression is: In the formula: W f b is the weight matrix of the forget gate; f This is the bias vector for the forget gate; 2) Input gate: The input gate consists of two parts. The first part uses the sigmoid activation function σ, and the output is i. t The first part takes values ​​in the range [0,1]; the second part uses the tanh activation function, and the output is... Its expression is: In the formula: W i b is the weight matrix of the input gate, used to control which new information should be updated to the cell state; i W is the bias vector of the input gate; C b is the weight matrix for candidate cell states, used to generate new candidate information; C is the bias vector for the candidate cell state; C represents the candidate cell state vector, with values ​​in the interval [-1, 1]. t-1 This represents the cell state at the previous time step, carrying long-term memory information; C t The updated current cell state combines the results of the forget gate and the input gate. 3) Output gate: The output gate controls how many cell states in this layer are filtered; first, the sigmoid activation function is used to obtain an o value in the range [0,1]. t Next, the cell state C t After processing with the tanh activation function, and then with o t Multiplication is the output h of this layer. t Its expression is: In the formula: W o b is the weight matrix of the output gate, used to control how much information about the cell state is output to the hidden state; o The bias vector of the output gate; o t h is the output vector of the output gate, with values ​​in the interval [0,1]. t The hidden state at the current time step is the output of the LSTM, which is passed to the next time step or used for prediction. The prediction result is as follows: y t =σ(W y ·h t +b y ) (15) In the formula: W y b is the weight matrix of the output layer; y y is the bias vector of the output layer; t This is the prediction output for the current time step.

6. A distribution network reliability assessment method based on EMD-PCA-LSTM, characterized in that, Includes the following steps: Step 1: Complete and clean the collected power grid operation data and environmental data that may be related to the reliability of the distribution network to construct the original dataset; Step 2: Decompose the original time series data into multiple intrinsic mode functions (IMFs) and a residual term through EMD decomposition; Step 3: Use PCA to reduce the dimensionality and extract features from the multiple IMF components obtained in Step 2, and retain the top k principal components with a cumulative variance contribution rate greater than 85% as the core features affecting the reliability of the distribution network. Step 4: Using the dimensionality reduction data obtained in Step 3, construct the final supervised learning dataset for LSTM, and divide it into training set, validation set, and test set; Step 5: Initialize the LSTM network, input the training set data, update the weights through backpropagation, and adjust the hyperparameters using the validation set; Step 6: Reliability assessment and prediction. Use the test set to evaluate the model performance and output reliability assessment metrics, including average outage duration (SAIDI) and average outage frequency (SAIFI).

7. A distribution network reliability assessment system based on EMD-PCA-LSTM, characterized in that, The aforementioned distribution network reliability assessment system based on EMD-PCA-LSTM is applied to the distribution network reliability assessment method based on EMD-PCA-LSTM as described in any one of claims 1-6.

Images