Short-term reliability evaluation method and system for power distribution network containing distributed power supply

Abstract

The invention discloses a short-term reliability evaluation method and system for a power distribution network containing a distributed power supply, and relates to the technical field of power grids. Obtaining historical wind power and photovoltaic output data and corresponding reliability index data based on the power distribution network reliability evaluation model, and taking the historical wind power and photovoltaic output data and the corresponding reliability index data as training samples of the hybrid deep learning model; historical wind power and photovoltaic output data are preprocessed, time seasonal information is extracted, and position coding is executed; a TCN-Wpsform hybrid deep learning model is constructed and trained, and the TCN-Wpsform hybrid deep learning model is constructed; and inputting to-be-evaluated wind power and photovoltaic output data into the trained hybrid deep learning model, and outputting a short-term reliability evaluation result of the power distribution network. According to the method, the complex nonlinear relationship between the distributed energy output and the reliability index is effectively captured, and the evaluation accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power grids, and in particular to a method and system for evaluating the short-term reliability of a power distribution network containing distributed power sources. Background Art

[0002] The existing reliability assessment methods for distribution networks with distributed generation have the problem of low computational efficiency. Although the traditional Monte Carlo simulation method can handle the randomness of the system, it requires a large amount of sampling calculations when dealing with large-scale distribution networks, which is time-consuming and difficult to meet the real-time requirements of short-term reliability assessment. When considering multiple fault combinations, the computational complexity of the state enumeration method increases exponentially as the system scale increases, which seriously limits its feasibility in practical applications.

[0003] Another significant shortcoming is that existing methods do not fully consider the characteristics of distributed power sources. Most evaluation methods fail to fully consider the intermittency, volatility, and correlation between distributed power sources and loads. At the same time, the impact of weather factors on the output of distributed power sources is not sufficiently considered, resulting in a large deviation between the evaluation results and the actual situation. In addition, existing methods often ignore the ability of distributed power sources to participate in system scheduling and fault recovery, and cannot accurately reflect their actual contribution to system reliability, thus affecting the accuracy and credibility of the evaluation results. Summary of the invention

[0004] In view of the problems existing in the prior art in terms of computational efficiency and model accuracy, the present invention is proposed.

[0005] Therefore, the problem to be solved by the present invention is how to significantly improve the calculation efficiency of reliability evaluation while ensuring accuracy.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In the first aspect, an embodiment of the present invention provides a short-term reliability assessment method for a distribution network containing distributed power sources, which includes establishing a distribution network reliability assessment model, determining reliability assessment indicators, formulating an island power supply recovery strategy, and setting load shedding rules corresponding to the island; based on the distribution network reliability assessment model, obtaining historical wind power and photovoltaic output data and their corresponding reliability index data as training samples for a hybrid deep learning model; preprocessing the historical wind power and photovoltaic output data, extracting time seasonal information and performing position encoding; constructing and training a TCN-Wpsformer hybrid deep learning model, wherein the TCN-Wpsformer hybrid deep learning model includes a temporal convolutional neural network module and a window probability sparse multi-head self-attention mechanism module; inputting the wind power and photovoltaic output data to be evaluated into the trained hybrid deep learning model, and outputting the short-term reliability assessment result of the distribution network.

[0008] As a preferred solution of the short-term reliability evaluation method of the distribution network containing distributed power sources described in the present invention, the reliability evaluation index includes the average power outage frequency index S AIFI , System average power outage time index S AIDI , Average system availability index A SAI and the expected power shortage E ENS , the specific formula is as follows:

[0009]

[0010]

[0011]

[0012]

[0013] Among them, N ld is the load node set, N Ck is the number of users at node k, L bk is the load demand of node k at load level b, Π bk is the expected outage rate of node k, Γ bk is the expected outage time of node k at load level b, B B is the load level set, Δ b is the average duration of b-level load, b is the load level, and k is the node.

[0014] As a preferred scheme of the short-term reliability assessment method of the distribution network containing distributed power sources described in the present invention, wherein: the load shedding rules corresponding to the island include rules 1 to 5, wherein rule 1 does not include DG and transfer points, rule 2 only includes DG, rule 3 only includes transfer points, rule 4 includes DG and transfer points and the transfer points are upstream of DG, and rule 5 includes DG and transfer points and the transfer points are downstream of DG.

[0015] As a preferred solution of the short-term reliability assessment method of the distribution network containing distributed power sources described in the present invention, the objective function of the distribution network reliability assessment model is to minimize the expected power shortage of the system, and the specific formula is as follows:

[0016] min E ENS

[0017] Among them, E ENS The expected power shortage of the system.

[0018] The constraints of the distribution network reliability assessment model include branch connectivity constraints, power supply restoration constraints, and interruption time constraints. The specific formula of the power supply restoration constraint is as follows:

[0019]

[0020]

[0021] in, is a 0-1 variable, N DG,z-down is the set of DG downstream load points of the feeder where branch z is located, is the set of load points downstream of the transfer point of the feeder where branch z is located, D Gz is the maximum available power of DG on the feeder where branch z is located, B z is the branch set of the feeder where branch z is located.

[0022] As a preferred solution of the method for short-term reliability assessment of a distribution network containing distributed power sources described in the present invention, the specific formula for executing position encoding is as follows:

[0023]

[0024]

[0025] Among them, j∈{1,...,[d model / 2]},d model is the preset data dimension, s is the absolute position information of the sequence where the data point is located, L is the length of the time series after data preprocessing and segmentation, P is the absolute position encoding data after processing, and P (s,2j) and P (s,2j+1) They are the 2j-dimensional and 2j+1-dimensional position values ​​of the s-th data respectively.

[0026] As a preferred solution of the short-term reliability assessment method of the distribution network containing distributed power sources described in the present invention, wherein: constructing and training the TCN-Wpsformer hybrid deep learning model includes the following steps: using the TCN network to extract time segment features from long-time series data of wind power and photovoltaic output, the TCN network includes a causal convolution layer and a dilated convolution layer; inputting the time segment features extracted by the TCN network into the Wpsformer model as input data of the window probability sparse multi-head self-attention mechanism; based on the input data, constructing a window sparse self-attention matrix and a probability sparse self-attention matrix in the Wpsformer model; splicing the window sparse self-attention matrix with the probability sparse self-attention matrix to obtain a window probability sparse self-attention mechanism model; performing layer normalization on the window probability sparse self-attention matrix, and adopting a residual connection strategy; inputting the window probability sparse self-attention matrix after layer normalization into a feedforward neural network for nonlinear transformation and feature fusion; the output of the feedforward neural network passes through a fully connected layer to generate the final distribution network short-term reliability assessment result.

[0027] As a preferred solution of the method for short-term reliability assessment of a distribution network containing distributed power sources described in the present invention, the specific formula of the window probability sparse self-attention mechanism model is as follows:

[0028]

[0029] in, is the window sparse self-attention matrix, is the non-overlapping part of the probabilistic sparse self-attention matrix and the window sparse self-attention matrix, Q is the query matrix, K is the key matrix, V is the value matrix, and softmax(·) is a fixed normalized exponential function.

[0030] In the second aspect, an embodiment of the present invention provides a short-term reliability assessment system for a distribution network containing distributed power sources, which includes an assessment module for establishing a distribution network reliability assessment model, determining reliability assessment indicators, formulating island power supply recovery strategies, and setting load shedding rules corresponding to islands; a sample acquisition module for acquiring historical wind power and photovoltaic output data and their corresponding reliability index data based on the distribution network reliability assessment model as training samples for a hybrid deep learning model; a preprocessing module for preprocessing historical wind power and photovoltaic output data, extracting temporal seasonal information and performing position encoding; a model construction module for constructing and training a TCN-Wpsformer hybrid deep learning model, wherein the TCN-Wpsformer hybrid deep learning model includes a temporal convolutional neural network module and a window probability sparse multi-head self-attention mechanism module; a generation module for inputting the wind power and photovoltaic output data to be evaluated into the trained hybrid deep learning model, and outputting the short-term reliability assessment result of the distribution network.

[0031] In a third aspect, an embodiment of the present invention provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program instructions are executed by the processor, the steps of the method for short-term reliability assessment of a distribution network containing distributed power sources as described in the first aspect of the present invention are implemented.

[0032] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program instructions are executed by a processor, the steps of the method for short-term reliability assessment of a distribution network containing distributed power sources as described in the first aspect of the present invention are implemented.

[0033] The beneficial effects of the present invention are as follows: the present invention effectively captures the complex nonlinear relationship between distributed energy output and reliability indicators, and improves the accuracy of the assessment. The introduced window probability sparse multi-head self-attention mechanism greatly reduces the computational complexity while maintaining good accuracy, so that the method can quickly process long time series data. This data-driven method avoids the traditional cumbersome manual modeling process and provides strong support for real-time reliability assessment and decision-making. In addition, by fully considering temporal seasonal information and position encoding, the data expression ability is enhanced, so that the model can better capture the temporal characteristics of wind power and photovoltaic output, further improving the accuracy and reliability of the evaluation results. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0035] Figure 1 The flowchart of the method for short-term reliability assessment of distribution network containing distributed generation.

[0036] Figure 2 The short-term reliability assessment method for the distribution network containing distributed generation does not include DG and transfer points in the island.

[0037] Figure 3 The short-term reliability assessment method for distribution network containing distributed generation only contains DG in the island.

[0038] Figure 4 The short-term reliability assessment method for the distribution network containing distributed generation only contains the transfer point in the isolated island.

[0039] Figure 5 The short-term reliability assessment method of the distribution network containing distributed generation is proposed in which the island contains DG and there is a transfer point upstream.

[0040] Figure 6 The short-term reliability assessment method of the distribution network containing distributed generation is proposed in which the island contains DG and there is a transfer point downstream.

[0041] Figure 7 The wind power data, time seasonal information and location coding are spliced ​​for the short-term reliability assessment method of distribution networks containing distributed generation.

[0042] Figure 8 It is a TCN-Wpsformer hybrid model for short-term reliability assessment of distribution networks containing distributed generation.

[0043] Fig. 9A temporal convolutional neural network is used to develop a short-term reliability assessment method for distribution networks containing distributed generation.

[0044] Fig.10 A window sparse self-attention mechanism model is proposed for the short-term reliability assessment method of distribution networks containing distributed generation.

[0045] Fig.11 A probabilistic sparse self-attention mechanism model for short-term reliability assessment of distribution networks containing distributed generation sources is proposed.

[0046] Fig.12 A window probability sparse self-attention mechanism model is proposed for the short-term reliability assessment method of distribution networks containing distributed generation.

[0047] Fig.13 The structure of the distribution system is a short-term reliability assessment method for the distribution network containing distributed generation. DETAILED DESCRIPTION

[0048] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0049] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0050] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0051] Example 1

[0052] Reference Figures 1 to 13 , which is the first embodiment of the present invention, and provides a method for evaluating the short-term reliability of a distribution network containing distributed power sources, comprising:

[0053] S1: Establish a distribution network reliability assessment model.

[0054] Preferably, the reliability optimization of the distribution network includes defining system characteristics and assumptions, determining reliability evaluation indicators, formulating island power supply restoration strategies, setting load shedding rules corresponding to the islands, and establishing a distribution network reliability evaluation model.

[0055] It should be noted that the distribution network short-term reliability assessment method proposed in the present invention is applicable to distribution networks with the following characteristics: each feeder contains at most one distributed generation node and one transfer point, equipped with a circuit breaker without reclosing, and disconnectors are installed at both ends of the branch; at the same time, it is based on the following assumptions: the tie line is completely reliable and the transmission power is unlimited, and the voltage constraint can be met; the disconnector operation time can be ignored. These characteristics and assumptions form the basis of the assessment method, simplify the model complexity, and ensure the practicality of the assessment results.

[0056] Furthermore, an island power supply restoration strategy is formulated. After any fault in the distribution network is isolated and reconstructed, except for the normal operation of the part connected to the main grid, only the island containing the distributed power source has the possibility of restoring power supply, and its power supply restoration strategy depends on the power abundance and network operation rules in the island. From a practical point of view, considering the complexity of the power flow backhaul management of the distribution network, such as protection configuration, personal safety of maintenance personnel, etc., the present invention assumes that the prescribed power flow direction of any branch before and after the fault is not allowed to change. Based on the above assumptions, the power supply mode in the island considered by the present invention is different from the traditional sufficiency assessment, which is specifically embodied in that the upstream loads of the power supply points (including transfer points and DG) in the island are not restored to power, and only the downstream loads are considered to be restored according to the maximum power supply.

[0057] Furthermore, load shedding rules corresponding to the island are set, including rules 1 to 5, where rule 1 does not include DG and transfer points, rule 2 only includes DG, rule 3 only includes transfer points, rule 4 includes DG and transfer points and the transfer points are upstream of DG, and rule 5 includes DG and transfer points and the transfer points are downstream of DG.

[0058] Specifically, Rule 1 (excluding DG and islands of transfer points) is as follows Figure 2 As shown in Figure 2, all loads on the island (the part shown by the dotted box) are powered off. If branches 5 to 6 fail, node 6 forms an island. At this time, nodes 2, DG, 4, and 5 only need to isolate the fault to restore power supply, while node 6 can only restore power supply after the fault is completely repaired. Rule 2 (island containing only DG) Figure 3 As shown in Figure 1, the load upstream of the DG node (such as node 4) does not participate in power supply restoration. The DG node and downstream nodes (such as node 6) selectively restore power supply based on the available output of the DG, load demand and optimization objectives. Rule 3 (only islands containing transfer points) is as follows Figure 4 As shown in , for a fault that occurs upstream of a transfer point (such as node 5): the load between the transfer point and the fault location (such as node 4) is not considered for power restoration, and the transfer point and its downstream loads (such as node 6) are considered for power restoration. Rule 4 (islands containing DG and transfer points, and the transfer point is upstream of DG) is as follows Figure 5As shown in Figure 2, when a fault occurs in branch 1-2, the upstream node of the transfer point (such as node 2) does not participate in the power supply restoration, and the load nodes between the transfer point and the DG node (excluding) (such as nodes 3 and 4) restore power supply. The power supply restoration of the DG node and its downstream nodes (such as node 6) is the same as Rule 2 (note that it is assumed that the transfer point and DG output can be adjusted freely to ensure that there will be no power flow reverse between node 3 and DG node). Rule 5 (including DG and transfer point, and the transfer point is an island downstream of DG) is as follows Figure 6 As shown in the figure, when branches 1-2 fail, the transfer point and its downstream loads resume power supply, the loads upstream of the DG node (such as node 2) do not participate in the power supply restoration, the power supply restoration of the load nodes between the DG node (inclusive) and the transfer point (excluding) is the same as rule 2, and all power supply is restored downstream of the transfer point.

[0059] Furthermore, the reliability evaluation index includes the average power outage frequency index (S AIFI ), system average power outage time index (S AIDI ), average system availability index (A SAI ) and the expected power shortage (E ENS ), the specific formula is as follows:

[0060]

[0061]

[0062]

[0063]

[0064] Among them, N ld is the load node set, N Ck is the number of users at node k, L bk is the load demand of node k at load level b, Π bk is the expected outage rate of node k, Γ bk is the expected outage time of node k at load level b, B B is the load level set, Δ b is the average duration of load level b, b is the load level, k is the node, S AIFI , S AIDI , A SAI 、E ENS They are the average power outage frequency index, the system average power outage time index, the average system availability index and the expected power shortage.

[0065] Furthermore, for the active distribution network with the above characteristics, the present invention proposes a reliability assessment method based on an assessment model. The core of the method is an optimization problem, including an objective function and a series of constraints. The objective function is to minimize the expected power shortage of the system. The specific formula is as follows:

[0066] min E ENS

[0067] Among them, E ENS The expected power shortage of the system.

[0068] Specifically, the constraints include branch connectivity constraints, power supply restoration constraints, and interruption time constraints. The specific formula for the branch connectivity constraint is as follows:

[0069]

[0070]

[0071]

[0072]

[0073] Among them, z∈B is a branch, B is a branch set, k∈N Ω is the node, N Ω is the node set, N z is the set of feeder load nodes where branch z is located, The indicator function indicating whether the node of the fault branch z is connected to the substation, The indicator function indicating whether the node of the fault branch z is connected to the DG, The indicator function that indicates whether the node of the fault branch z is connected to the transfer point, The meaning is: if branch z fails, the first section circuit breaker of the feeder where the branch is located will be activated, and all load points on this feeder will be out of power.

[0074] Preferably, the specific formula of the power supply restoration constraint is as follows:

[0075]

[0076]

[0077] in, is a 0-1 variable, N DG,z-down is the set of load points downstream of the DG in the feeder where branch z is located (when there is a transfer point downstream of the DG, N DG,z-down is the set of load nodes from DG to the transfer point, excluding the transfer point), is the set of load points downstream of the transfer point of the feeder where branch z is located (when there is a DG downstream of the transfer point, is the load node set from the transfer point to DG, excluding DG), D Gz is the maximum available power of DG on the feeder where branch z is located, B z is the branch set of the feeder where branch z is located.

[0078] It should be noted that The meaning is: if branch z fails and the network is reconfigured, the upstream load node of branch z will not resume power supply temporarily, that is, All load nodes outside the feeder where branch z is located can resume power supply, that is, The meaning is: if branch z fails, the first section circuit breaker of the feeder where the branch is located will be activated, and all load points on the feeder will be out of power. And because the power supply restoration constraint is a nonlinear constraint, the large M method can be used to linearize it.

[0079] Furthermore, the island formed by the downstream node of the feeder where branch z is located can be divided into the following cases: The island does not contain DG and transfer point: All load nodes are powered off, that is, The island only contains DG or transfer point: To avoid power flow reversal, the load nodes upstream of DG and transfer point Power supply is not restored temporarily, DG downstream load node k∈N DG,z-down Through optimization, selective power restoration is performed to the load nodes downstream of the transfer point. All power supply is restored. The island contains both DG and transfer point: if DG is downstream of the transfer point, the power supply restoration of the island load node is the same as load shedding rule 4; if DG is upstream of the transfer point, the power supply restoration of the island load node is the same as load shedding rule 5.

[0080] It should be noted that the power supply restoration constraint can constrain the power supply restoration result of node k That is: for have:

[0081]

[0082] in, is a 0-1 variable. When the load point k and the fault branch z are not on the same feeder, The node is not affected by the faulty branch and still maintains continuous power supply, that is, When the load point k and the fault branch z are on the same feeder, The feeder head end circuit breaker trips, and its power supply restoration depends on the connectivity and power abundance with the DG or transfer point, that is,

[0083] Furthermore, for the interruption time constraint, for any node k, the expected interruption rate Π under load level b is introduced: bk and expected interruption time Γ bkTwo parameters, let N s is the substation node set. Since substation failure is not considered, have:

[0084]

[0085] And for have:

[0086]

[0087] Among them, λ z , They represent the failure rate, isolation time and isolation repair time of branch z respectively. is a 0-1 variable, z is a branch, Π bk is the expected outage rate, Γ bk is the expected interruption time, The meaning is: if branch z fails, the first section circuit breaker of the feeder where the branch is located will be activated, and all load points on the feeder will be powered off. ld is the set of load nodes.

[0088] Through this evaluation model, the optimal power supply restoration plan under various fault conditions can be obtained, and then the reliability indicators of the distribution network can be evaluated.

[0089] S2: Based on the distribution network reliability assessment model, historical wind power and photovoltaic output data and their corresponding reliability index data are obtained as training samples for the hybrid deep learning model.

[0090] Specifically, historical wind power and photovoltaic output data are collected, which reflect the actual operation of distributed energy; the distribution network reliability assessment model established in S1 is used to calculate the corresponding reliability indicators based on the collected historical data, including S AIFI , S AIDI , A SAI and E ENS ; Pair historical wind power and photovoltaic output data with their corresponding reliability index data to form a training sample set for the hybrid deep learning model.

[0091] It should be noted that the reliability assessment of the distribution network usually assumes that the output power of the main power source is unlimited, so the reliability of the distribution network is mainly affected by the output of distributed wind power and photovoltaic power. By adopting a deep learning method, the present invention aims to explore the intrinsic relationship between the output of distributed energy and the reliability index of the distribution network. This method can accurately capture the complex nonlinear relationship between wind power, photovoltaic output and reliability index. After sufficient training, it can achieve a rapid assessment of the short-term reliability of the distribution network. This data-driven method can not only improve the accuracy of the assessment, but also significantly reduce the calculation time, providing strong support for the real-time reliability assessment and decision-making of the distribution network.

[0092] S3: Preprocess historical wind power and photovoltaic output data, extract temporal seasonal information and perform position encoding.

[0093] Specifically, this step aims to preprocess the historical wind power and photovoltaic output data and introduce temporal seasonal information and position encoding to enhance the data expression ability. In order to enable the Transformer neural network to capture the temporal position characteristics, absolute position encoding is introduced, and the encoding formula is:

[0094]

[0095]

[0096] Among them, j∈{1,...,[d model / 2]},d model is the preset data dimension, s is the absolute position information of the sequence where the data point is located, L is the length of the time series after data preprocessing and segmentation, P is the absolute position encoding data after processing, and P (s,2j) and P (s,2j+1) They are the 2j-dimensional and 2j+1-dimensional position values ​​of the s-th data respectively (2j and 2j+1 represent the parity of the trigonometric function).

[0097] When processing time series data, in addition to direct location information, time hidden features are equally important. In order to fully capture time hidden features, global time features are extracted from each timestamp, including time code, hour code, date code, and month code. These codes are normalized according to their respective dimensions and then spliced ​​together to fully reflect the time characteristics of time series data.

[0098] Specifically, Figure 7 As shown, u n is the distributed wind and solar output data series, p n is the absolute position encoding data, t n is the moment encoding data, h n is the whole point coded data, d nFor date coded data, m n The month coding data is concatenated with the original data, position coding and time feature coding to obtain the final input data c n , the specific formula is as follows:

[0099]

[0100] in, Represents the i-th column data of the t-th sequence input, represents the normalized vector of the original data of distributed wind and solar output in the i-th column of the t-th sequence input, P i t Represents the absolute position encoding vector of the i-th column data of the t-th sequence input, represents the encoding vector of the i-th column data of the t-th sequence input normalized according to the d-time dimension, α is the balance factor, The matrix consisting of the input data for the nth sequence.

[0101] S4: Build and train the TCN-Wpsformer hybrid deep learning model.

[0102] Specifically, the network structure of the TCN-Wpsformer hybrid deep learning model is as follows Figure 8 As shown, the following steps are included:

[0103] S4.1: Use the TCN network to extract time segment features from long-term series data of wind power and photovoltaic output. The TCN network includes a causal convolution layer and a dilated convolution layer.

[0104] Specifically, the TCN network has made innovations based on the CNN model, mainly in two aspects: causal convolution and dilated convolution. Fig. 9 As shown in the figure, causal convolution ensures that when extracting the features at the sk moment, the feature value only depends on the data at the current moment and before, avoiding the "information leakage" problem. Dilated convolution performs interval sampling during the convolution process, so that the size of the receptive field increases exponentially with the number of network layers, thereby obtaining a larger receptive field while maintaining a small number of layers. The specific formula of the TCN network is as follows:

[0105]

[0106] Among them, H(sk) is the output of the tth neuron in the TCN network at time sk, h(s) is the sth element in the convolution kernel, and x sk-dm·s is the input value corresponding to the moment dm·s shifted from moment sk to the past, dm is the dilation factor, k is the convolution kernel size, and sk is the moment.

[0107] It should be noted that when dm=1, the dilated causal convolution is transformed into the standard causal convolution, and the effective historical information that can be extracted by one layer is dm·(k-1).

[0108] Furthermore, the introduction of the TCN network is optimized for the special needs of short-term reliability assessment. Specifically, considering the dynamic impact of extreme weather, day and night and seasonal changes on wind power, photovoltaic output characteristics and distribution network reliability, the TCN network expands the model receptive field by increasing the number of network layers, thereby more effectively extracting time segment feature information. In addition, the introduction of the TCN network also overcomes the inherent limitations of the Transformer-like model. Compared with the traditional Transformer model where query, key and value are only composed of c n Vector and weight matrix W Q , W K , W V The point product is different. The TCN network changes the self-attention mechanism structure so that the query and key inputs contain feature information of different time segments, while the value remains the original time series data input. This structural design enables the TCN network to more accurately calculate the correlation between the query and the key, effectively establish the connection between each time segment, and capture the feature information of important historical time segments, thereby improving the accuracy of short-term reliability assessment.

[0109] S4.2: The time segment features extracted by the TCN network are input into the Wpsformer model as the input data of the windowed probabilistic sparse multi-head self-attention mechanism.

[0110] Specifically, the distributed wind and solar output data after TCN processing are further input into the Wpsformer model. When processing long time series data, the multi-head attention mechanism in the traditional Transformer model often faces the challenges of high computational complexity and slow computational speed due to the long sequence length. To this end, the present invention proposes a window probability sparse multi-head self-attention mechanism. In essence, the window probability sparse multi-head self-attention mechanism combines the local attention mechanism with a fixed window size value and the probabilistic sparse self-attention mechanism for relatively important query vectors. When facing the processing of long time series data, it reduces the computational complexity while still maintaining good computational accuracy.

[0111] Specifically, the specific formula of the original Transformer's multi-head self-attention is as follows:

[0112]

[0113] Among them, A(·) is the attention function, softmax(·) is a fixed normalized exponential function, Q is the query matrix, K is the key matrix, V is the value matrix, and d k is the input dimension and output dimension at each moment.

[0114] Furthermore, the specific formula of the multi-head self-attention mechanism is as follows:

[0115] M(Q,K,V)=Concat(H1,...,H d )W o

[0116] Among them, M(·) is the multi-head self-attention function, H d is the d-th self-attention function, Concat(·) is the concatenation function, W o is the linear mapping matrix of the multi-head self-attention function, Q is the query matrix, K is the key matrix, and V is the value matrix.

[0117] Furthermore, the specific formula of the function structure of the d-th self-attention in the multi-head self-attention mechanism is as follows:

[0118]

[0119] Among them, QW d Q , KW d K 、VW d V is the linear mapping matrix of Q, K, V in the d-th self-attention function, A(·) is the attention function, H d is the d-th self-attention function.

[0120] S4.3: Based on the input data, a window sparse self-attention matrix is ​​constructed in the Wpsformer model, and local attention calculation is performed on the data in a fixed window at each observation point.

[0121] Specifically, the attention matrix structure of the window sparse self-attention mechanism is as follows Fig.10 As shown. In the short-term reliability assessment of the distribution network, data closer to the prediction point in a short time scale is often more valuable than observation points farther away from the prediction point. Unlike the original Transformer model, which needs to calculate the correlation between any two observation points, the window sparse self-attention mechanism uses a local attention mechanism with a fixed window size to only pay attention to and process a fixed number of observation points near each observation point. Fig. 9In the self-attention matrix, each observation point only uses the local attention mechanism to process the observation points near the fixed window size, and the white part outside the window is filled with zero vectors. n represents the length of the input sequence, and w represents the set window size. The computational complexity of the original Transformer model is O(n 2 ), as the sequence length n increases, its computational complexity will increase exponentially. In contrast, the computational complexity of the window sparse self-attention mechanism is only O(nw), which greatly improves the computational efficiency.

[0122] Furthermore, in the window sparse self-attention mechanism, when the number of layers of the mechanism is l and the window size w is fixed, the receptive field of the top layer will reach lw. Fig.10 The right side of is intuitively displayed: in the first layer, the attention coverage of observation point a2 is from a1 to a3, and its window size is w; while in the second layer, the attention coverage of observation point b2 is from b1 to b3, and the window size is also w. It is worth noting that b1 in the second layer already includes the local information captured by a2 in the first layer. Similarly, b3 also covers the information captured by a4. Therefore, b2 in the second layer can actually utilize all the information from a1 to a5. This mechanism means that although the bottom layer can only capture limited information, as the layers increase, the receptive field of the top layer will gradually widen and integrate more information.

[0123] S4.4: Based on the input data, a probabilistic sparse self-attention matrix is ​​constructed in the Wpsformer model, and the query vector and key vector with higher importance are selected for dot product operation.

[0124] Specifically, the self-attention matrix of the probabilistic sparse self-attention mechanism is as follows Fig.11 As shown. From the above analysis, it can be seen that the window size value of each layer in the window sparse self-attention mechanism is fixed, so it is impossible to flexibly perform adaptive adjustments according to the changes in time series data. In addition, in addition to the short-term features of the time segments adjacent to the observation point, time series with higher importance should be considered. In order to make up for the shortcomings of the window sparse self-attention mechanism, the present invention proposes a probabilistic sparse self-attention mechanism. In fact, when calculating self-attention, only the dot product of a few feature data contributes significantly to the main attention, while the influence of other dot products is relatively weak and can be ignored. Therefore, the probabilistic sparse self-attention mechanism will first sort the query vectors according to importance, screen out the more important query vectors, and then perform dot product calculations. In this way, each key vector will only focus on o dominant query vectors, thereby constructing a probabilistic sparse self-attention matrix. Fig.11The dark part in the middle represents the dot product operation between the o dominant query vectors and each key vector to calculate the correlation between the observation points, and the remaining white part is filled with zero vectors to avoid unnecessary updates.

[0125] S4.5: Concatenate the window sparse self-attention matrix with the probability sparse self-attention matrix to obtain the window probability sparse self-attention mechanism model.

[0126] Preferably, the specific formula of the window probability sparse self-attention mechanism model is as follows:

[0127]

[0128] in, is the window sparse self-attention matrix, is the non-overlapping part of the probability sparse self-attention matrix and the window sparse self-attention matrix, that is Fig.11 The orange part.

[0129] S4.6: Perform layer normalization on the window probability sparse self-attention matrix and adopt a residual connection strategy.

[0130] Furthermore, the window probability sparse self-attention matrix is ​​processed through layer normalization technology to stabilize the internal feature distribution, and combined with the residual connection strategy, it can alleviate the gradient vanishing and network performance degradation problems that may be caused by the increase in the number of layers of the deep neural network, and improve the convergence speed and stability of the reliability assessment results.

[0131] S4.7: The window probability sparse self-attention matrix after layer normalization is input into the feedforward neural network for nonlinear transformation and feature fusion.

[0132] Specifically, the specific formulas for nonlinear transformation and feature fusion are as follows:

[0133] F(k)=max(0,kW1+b1)W2+b2

[0134] Where F(·) is the feedforward neural network function, W1 and W2 are the weight matrices of the first and second linear layers, b1 and b2 are the bias vectors of the first and second linear layers, and k is the output of the normalization layer.

[0135] S4.8: The output of the feedforward neural network passes through the fully connected layer to generate the final distribution network short-term reliability assessment results.

[0136] Finally, the TCN-Wpsformer model outputs the results of short-term reliability assessment of the distribution network through the fully connected layer.

[0137] S5: Input the wind power and photovoltaic output data to be evaluated into the trained hybrid deep learning model, and output the short-term reliability evaluation results of the distribution network.

[0138] Furthermore, the present embodiment also provides a short-term reliability assessment system for a distribution network containing distributed power sources, including an assessment module for establishing a distribution network reliability assessment model, determining reliability assessment indicators, formulating an island power supply recovery strategy, and setting load shedding rules corresponding to the island; a sample acquisition module for acquiring historical wind power and photovoltaic output data and their corresponding reliability indicator data based on the distribution network reliability assessment model as training samples for a hybrid deep learning model; a preprocessing module for preprocessing historical wind power and photovoltaic output data, extracting temporal seasonal information and performing position encoding; a model construction module for constructing and training a TCN-Wpsformer hybrid deep learning model, wherein the TCN-Wpsformer hybrid deep learning model includes a temporal convolutional neural network module and a window probability sparse multi-head self-attention mechanism module; and a generation module for inputting the wind power and photovoltaic output data to be evaluated into the trained hybrid deep learning model, and outputting the short-term reliability assessment result of the distribution network.

[0139] This embodiment also provides a computer device, which is suitable for the case of a short-term reliability assessment method for a distribution network containing distributed power sources, including a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the short-term reliability assessment method for a distribution network containing distributed power sources as proposed in the above embodiment.

[0140] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0141] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the method for short-term reliability assessment of a distribution network containing distributed power sources proposed in the above embodiment is implemented.

[0142] In summary, the present invention effectively captures the complex nonlinear relationship between distributed energy output and reliability indicators, and improves the accuracy of the assessment. The introduced window probability sparse multi-head self-attention mechanism greatly reduces the computational complexity while maintaining good accuracy, enabling the method to quickly process long time series data. This data-driven approach avoids the traditional cumbersome manual modeling process and provides strong support for real-time reliability assessment and decision-making. In addition, by fully considering temporal seasonal information and position encoding, the data expression capability is enhanced, enabling the model to better capture the temporal characteristics of wind power and photovoltaic output, further improving the accuracy and reliability of the assessment results.

[0143] Example 2

[0144] Reference Figures 1 to 13 , which is the second embodiment of the present invention, and this embodiment provides a short-term reliability assessment method for a distribution network containing distributed power sources. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0145] Specifically, the test system is IEEE RBTS BUS 6, and its structure diagram is as follows Fig.13 As shown. The feeder contains 23 load points, 1183 users, and a total user power of 33.63MW. The specific load data is shown in Table 1. The system is also connected to 10 1.5MW photovoltaic panels, whose output power obeys the Beta distribution with shape parameters of 2.06 and 2.5. The data set sampled based on this distribution is used as the historical data set. The power line failure rate is 0.05f / (a·km), and the repair time is 4h; the circuit breaker failure rate is 0.002f / a, and the repair time is 4h; the transformer failure rate is 0.015f / a, and the repair time is 200h; the switching time of the section switch is 1 minute.

[0146] Table 1 Load data

[0147] Number of load points Load point User Type Number of users Average Power 1 2 Residential 126 1.81 2 1，6 Residential 147 1.90 1 5 Residential 132 1.12 4 8，11，14，19 Residential 79 1.63 4 10，12，16，22 Residential 76 1.09 2 15，20 agriculture 1 1.27 3 3，13，17 agriculture 1 1.54 2 4，18 agriculture 1 1.15 2 7，23 agriculture 1 1.48 2 9，21 agriculture 1 1.80 ... ... ... ... ... total 23 ... 1183 33.63

[0148] Preferably, the present invention uses the proposed method to predict the reliability of the distribution network for 7*24 hours in the next week. The sample size N is set to 1000, and the sample ratio for training, verification and testing is 7:1:2. According to the output power of each sample distributed power source, the reliability evaluation method is used to calculate the actual reliability index corresponding to the sample.

[0149] Furthermore, the MAPE (Mean Absolute Percentage Error) of the proposed method and the traditional method are calculated respectively. The MAPE of each reliability index is shown in Table 2.

[0150] Table 2 MAPE of the present invention and the traditional method

[0151] ASAI AENS SAIFI SAIDI CAIDI The present invention 0.001％ 0.53％ 0.02％ 0.92％ 0.86％ Traditional methods 0.021％ 2.12％ 0.01％ 2.19％ 2.18％

[0152] Furthermore, through example analysis, it can be seen that the proposed method can significantly improve the accuracy of short-term reliability assessment of distribution network and can control the accuracy of short-term reliability assessment of distribution network within 1%.

[0153] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for short-term reliability assessment of a distribution network containing distributed power sources, characterized in that: include, Establish a distribution network reliability assessment model, determine reliability assessment indicators, formulate island power supply restoration strategies, and set load shedding rules corresponding to islands; Based on the distribution network reliability assessment model, historical wind power and photovoltaic output data and their corresponding reliability index data are obtained as training samples for the hybrid deep learning model; Preprocess historical wind power and photovoltaic output data, extract temporal seasonal information and perform position encoding; Construct and train a TCN-Wpsformer hybrid deep learning model, wherein the TCN-Wpsformer hybrid deep learning model includes a temporal convolutional neural network module and a windowed probabilistic sparse multi-head self-attention mechanism module; The wind power and photovoltaic output data to be evaluated are input into the trained hybrid deep learning model, and the short-term reliability evaluation results of the distribution network are output.

2. The method for short-term reliability assessment of a distribution network containing distributed power sources according to claim 1, characterized in that: The reliability evaluation index includes the average power outage frequency index S AIFI , System average power outage time index S AIDI , Average system availability index A SAI and the expected power shortage E ENS , the specific formula is as follows: Among them, N ld is the load node set, N Ck is the number of users at node k, L bk is the load demand of node k at load level b, Π bk is the expected outage rate of node k, Γ bk is the expected outage time of node k at load level b, B B is the load level set, Δ b is the average duration of b-level load, b is the load level, and k is the node.

3. The method for short-term reliability assessment of a distribution network containing distributed power sources according to claim 1, characterized in that: The load shedding rules corresponding to the island include rules 1 to 5, wherein rule 1 does not include DG and transfer points, rule 2 only includes DG, rule 3 only includes transfer points, rule 4 includes DG and transfer points and the transfer points are upstream of DG, and rule 5 includes DG and transfer points and the transfer points are downstream of DG.

4. The method for short-term reliability assessment of a distribution network containing distributed power sources according to claim 1, characterized in that: The objective function of the distribution network reliability assessment model is to minimize the expected power shortage of the system. The specific formula is as follows: my E ENS Among them, E ENS is the expected power shortage of the system; The constraints of the distribution network reliability assessment model include branch connectivity constraints, power supply restoration constraints and interruption time constraints. The specific formula of the power supply restoration constraint is as follows: in, is a 0-1 variable, N DG,z-down is the set of DG downstream load points of the feeder where branch z is located, is the set of load points downstream of the transfer point of the feeder where branch z is located, D Gz is the maximum available power of DG on the feeder where branch z is located, B z is the branch set of the feeder where branch z is located.

5. The method for short-term reliability assessment of a distribution network containing distributed power sources according to claim 1, characterized in that: The specific formula for executing position encoding is as follows: Among them, j∈{1,...,[d model / 2]},d model is the preset data dimension, s is the absolute position information of the sequence where the data point is located, L is the length of the time series after data preprocessing and segmentation, P is the absolute position encoding data after processing, and P (s,2j) and P (s,2j+1) They are the 2j-dimensional and 2j+1-dimensional position values ​​of the s-th data respectively.

6. The method for short-term reliability assessment of a distribution network containing distributed power sources according to claim 1, characterized in that: The construction and training of the TCN-Wpsformer hybrid deep learning model includes the following steps: The TCN network is used to extract time segment features from long-term series data of wind power and photovoltaic output. The TCN network includes a causal convolution layer and a dilated convolution layer. The time segment features extracted by the TCN network are input into the Wpsformer model as the input data of the windowed probabilistic sparse multi-head self-attention mechanism; Based on the input data, the window sparse self-attention matrix and the probability sparse self-attention matrix are constructed in the Wpsformer model; The window sparse self-attention matrix is ​​concatenated with the probability sparse self-attention matrix to obtain the window probability sparse self-attention mechanism model; The window probability sparse self-attention matrix is ​​layer-normalized and a residual connection strategy is adopted; The window probability sparse self-attention matrix after layer normalization is input into the feedforward neural network for nonlinear transformation and feature fusion; The output of the feedforward neural network passes through the fully connected layer to generate the final distribution network short-term reliability assessment results.

7. The method for short-term reliability assessment of a distribution network containing distributed power sources according to claim 6, characterized in that: The specific formula of the window probability sparse self-attention mechanism model is as follows: in, is the window sparse self-attention matrix, is the non-overlapping part of the probabilistic sparse self-attention matrix and the window sparse self-attention matrix, Q is the query matrix, K is the key matrix, V is the value matrix, and softmax(·) is a fixed normalized exponential function.

8. A short-term reliability assessment system for a distribution network containing distributed power sources, based on the short-term reliability assessment method for a distribution network containing distributed power sources according to any one of claims 1 to 7, characterized in that: Also includes, The evaluation module is used to establish a distribution network reliability evaluation model, determine reliability evaluation indicators, formulate island power supply restoration strategies, and set load shedding rules corresponding to islands; The sample acquisition module is used to obtain historical wind power and photovoltaic output data and their corresponding reliability index data based on the distribution network reliability assessment model as training samples for the hybrid deep learning model; The preprocessing module is used to preprocess the historical wind power and photovoltaic output data, extract the temporal seasonal information and perform position encoding; A model building module, used to build and train a TCN-Wpsformer hybrid deep learning model, wherein the TCN-Wpsformer hybrid deep learning model includes a temporal convolutional neural network module and a windowed probabilistic sparse multi-head self-attention mechanism module; The generation module is used to input the wind power and photovoltaic output data to be evaluated into the trained hybrid deep learning model and output the short-term reliability evaluation results of the distribution network.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method for short-term reliability assessment of a distribution network containing distributed power sources as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for short-term reliability assessment of a distribution network containing distributed power sources described in any one of claims 1 to 7 are implemented.

Images