A method for constructing a training data set

Abstract

The application provides a method for constructing a training data set, which is used for training an importance prediction model of a power system, the power system comprising a plurality of nodes and a plurality of lines, a plurality of power generation units and at least one load unit, and comprising the following steps: obtaining actual sampling data of the power system corresponding to a time sequence {t i} wherein i∈[1,T]; applying a disturbance to the power system based on a randomly selected scenario mode in a subsequence {t i} of the time sequence {t i′}, and calculating corresponding simulation sampling data; reconstructing the actual sampling data using the simulation sampling data to obtain reconstructed sampling data of the power system; and constructing the training data set based on the reconstructed sampling data. The training data set constructed by the method of the application can effectively improve the prediction accuracy of the prediction model for various emergency conditions when the training data set is used to train the importance prediction model of the power system.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of power system control, and specifically provides a construction method of a training data set of a power system importance prediction model. BACKGROUND

[0002] Safe and stable operation of a power system is an important guarantee for production, transportation and daily life of people. Analysis of a large number of power system faults shows that failure, withdrawal from operation, etc. of some important nodes or main lines often spreads to the entire power grid, thereby causing huge economic losses and even affecting social stability.

[0003] In order to minimize or eliminate the occurrence probability of the above faults, in addition to real-time monitoring of nodes and lines of the power system through various technical means, a scheme of predicting the importance of nodes and lines according to power flow data of the power system is proposed, striving to provide early warning capability of protection signals for the power grid. For example, patent 202111054797.X proposes a power system node importance prediction method, which can predict the importance of power system nodes in a future period of time by using historical power flow data of the power system.

[0004] However, in the actual operation process of the power system, due to the adoption of various real-time monitoring and intelligent peak shaving means to ensure the safety and stability of the power grid, the obtained historical power flow data often presents stable slight fluctuations or regular periodic tidal characteristics, resulting in that the prediction model only has the ability to predict the importance of the power system in a smooth operation, and cannot accurately predict the importance of the power system when the system power flow is subjected to a sudden condition leading to a sharp fluctuation, thereby greatly limiting the application scenarios and prediction accuracy of the prediction model. SUMMARY

[0005] To solve the problems existing in the prior art, the purpose of the present application is to optimize the training data set of the existing power system importance prediction model and its construction method, so as to improve the training effect of the power system importance prediction model.

[0006] The embodiments of the present application can be realized through the following technical solutions:

[0007] A construction method of a training data set, the training data set being used for training an importance prediction model of a power system, the power system comprising a plurality of nodes, a plurality of lines, a plurality of power generation units and at least one load unit, comprising the following steps:

[0008] S1: obtaining actual sampling data of nodes and lines of the power system corresponding to a time sequence {t i}i∈[1,T];

[0009] S2: Scenario patterns based on random selection in {t i The subsequence {t} of} i′ A disturbance is applied to the power system and the corresponding simulated sampling data is calculated. The simulated sampling data is then used to reconstruct the actual sampling data to obtain the reconstructed sampling data of the power system.

[0010] S3: Construct the training dataset based on the reconstructed sampled data.

[0011] Furthermore, the actual sampling data includes the data of each node, line, power generation unit, and load unit at each time point t. i The actual sampled values ​​of the corresponding active and reactive power.

[0012] Preferably, the scenario modes include a power failure mode and a line failure mode.

[0013] Further, step S2 includes the following steps:

[0014] S21: Randomly select at least one scene mode;

[0015] S22: If the selected scenario mode includes a power failure mode, then at least one power generation unit is further randomly selected and its time points t are set. i′ The corresponding active power disturbance ΔP G (t i′ and reactive power disturbance ΔQ G (t i′ ),as well as,

[0016] If the selected scenario mode includes a line failure mode, then at least one line is randomly selected and its time points t are set. i′ On / off state;

[0017] S23: Based on the aforementioned ΔP G (t i′ ), ΔQ G (t i′ ) and the change in on / off state determines the time series {t i′ The corresponding simulated sampling data includes nodes, lines, generation units, and load units at each time point t. i′ The corresponding simulated sampled values ​​of active and reactive power;

[0018] S24: The simulated sampling data is used to reconstruct the actual sampling data to obtain the reconstructed sampling data of the power system.

[0019] Further, the power supply fault mode is specifically that the active power and the reactive power of the failed power generation unit are reduced compared with the actual sampling values.

[0020] Further, the line fault mode is specifically that the active power and the reactive power of the failed line are reduced to 0.

[0021] Further, the calculation of the analog sampling data satisfies the node power balance constraint.

[0022] Further, the step S3 comprises the following steps:

[0023] S31: establishing a directed weighted graph of the power system;

[0024] S32: obtaining power flow data of the power system corresponding to the reconstructed sampling data time sequence {t i} based on the reconstructed sampling data;

[0025] S33: calculating the importance of each node at each time point t i based on the power flow data and the directed weighted graph;

[0026] S34: calculating the importance of each line at each time point t i based on the power flow data and the directed weighted graph;

[0027] S35: generating the training data set, the training data set comprising the importance of each node and each line at each time point t i .

[0028] Preferably, the importance of each node at each time point t i is determined based on an EB-SALSA algorithm.

[0029] Preferably, the importance of each line at each time point t i is determined based on an EL-SALSA algorithm.

[0030] The method for constructing a training data set provided by the embodiments of the present application has at least the following beneficial effects:

[0031] The technical solution of the present application is based on real sampling data of the power system, by simulating the changes of the states of each part of the system under different fault scenarios and superimposing them as disturbances on the real sampling data, so that the constructed training data set can reflect the changes of the node and line importance of the power system under various fault conditions. Using the training data set constructed by this method to train the power system importance prediction model can effectively increase the ability of the prediction model to identify the corresponding power flow evolution trend of the power system under various fault scenarios, and can make the prediction model more accurately predict the importance of the core node and the main line at the future time point when various emergencies occur. BRIEF DESCRIPTION OF DRAWINGS

[0032] Figure 1 The topology structure of an IEEE39 bus system according to one specific embodiment of the present application;

[0033] Figure 2 The topology structure of an IEEE118 bus system according to one specific embodiment of the present application;

[0034] Figure 3 The flowchart of a training data set construction method according to an embodiment of the present application;

[0035] Figure 4 The change and comparison results of the importance of a specific node according to embodiment 1 of the present application;

[0036] Figure 5 The change and comparison results of the importance of a specific line according to embodiment 1 of the present application;

[0037] Figure 6 The change and comparison results of the importance of a specific node according to embodiment 2 of the present application;

[0038] Figure 7 The change and comparison results of the importance of a specific line according to embodiment 2 of the present application. DETAILED DESCRIPTION

[0039] Hereinafter, the present application will be further described based on the preferred embodiments and with reference to the accompanying drawings.

[0040] In addition, in order to facilitate understanding, various components on the drawing are enlarged or reduced, but this practice is not intended to limit the protection scope of the present application.

[0041] The singular form of the word also includes the plural meaning, and vice versa.

[0042] In the description in the embodiments of the present application, it should be noted that if the terms "upper", "lower", "inner", "outer" and the like indicating the orientation or position relationship are based on the orientation or position relationship shown in the drawings or the orientation or position relationship in which the product of the embodiments of the present application is usually placed, and are only for the convenience of describing the present application and simplifying the description, and therefore cannot be understood as indicating or implying that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as limiting the present application. In addition, in the description of the present application, in order to distinguish different units, the first, second and the like are used in the specification, but these are not limited by the order of manufacture, and cannot be understood as indicating or implying relative importance, and the names may be different in the detailed description and claims of the present application.

[0043] The words in the specification are used to illustrate the embodiments of the present application, but are not intended to limit the present application. It should be noted that unless otherwise explicitly specified and limited, if the terms "provided", "connected", "connected" appear, they should be understood in a broad sense, for example, they can be fixedly connected, or can be detachably connected, or integrally connected; can be mechanically connected, can be directly connected, or indirectly connected through an intermediate medium, or can be connected inside two elements. For those skilled in the art, the specific meaning of the above terms in the present application can be specifically understood.

[0044] The present application provides a method for constructing a training data set, which is used to train an importance prediction model of a power system, wherein the power system includes a plurality of nodes, a plurality of lines, a plurality of power generation units and at least one load unit.

[0045] Figure 1 The topological structure of a specific power system is shown, which is an IEEE39 bus system, specifically including 10 power generation units (represented by circles with the letter "G" in the figure), 21 load units (represented by black arrows in the figure), 39 nodes (represented by black circles numbered 1-39 in the figure) and 46 lines.

[0046] Figure 2 The topological structure of another specific power system is shown, which is an IEEE118 bus system, specifically including 54 power generation units (represented by hollow circles in the figure), 99 load units (represented by black arrows in the figure), 118 nodes (represented by black circles in the figure) and 186 lines.

[0047] As Figure 1 , Figure 2In the shown power system, the power output by each power generation unit is input to each load unit through different nodes and lines, and the entire system satisfies the energy conservation. At any time point, the active and reactive power flowing out of the power generation unit, flowing through each node and each line, and finally flowing into the load unit is sampled, the sampling values of the active and reactive power of each node, line, power generation unit and load unit are obtained, and the system power flow data at that time is calculated based on the system power conservation condition. The method of calculating the system power flow data based on the power sampling values of each node, line and power generation and load unit of the power system is well known to those skilled in the art, for example, Newton-Raphson method can be used for iterative calculation to obtain convergent system power flow.

[0048] In the above-mentioned power system, the contribution and influence degree of each node and line to the entire system are different, and by applying additional and special protection to important nodes and lines, the stable operation of the system can be ensured. There are currently methods for calculating the importance of nodes and lines based on real-time power flow data of the power system using EB-SALSA algorithm and EL-SALSA algorithm respectively (Geng, J. Q. Research on important nodes and lines evaluation and backbone network planning of power system[D]. Harbin Institute of Technology, 2020.), and on this basis, patent 202111054797.X further proposes a power system node importance prediction method, which uses a convolutional long short-term memory network to establish a prediction model, and by using historical power flow data of the power system, the importance of the power system nodes in the future period of time can be predicted. After expanding the size of the memory network of the above-mentioned prediction model, a power system importance prediction model capable of simultaneously predicting the importance of nodes and lines of the power system can also be formed.

[0049] The performance of the above-mentioned prediction model, including the accuracy of the prediction and the robustness of the model, is closely related to the data set used for training in addition to being affected by the parameter settings of each layer inside the prediction model. Only when the training set covers various types of power grid flow characteristics, the trained prediction model can identify the characteristics of the power grid flow as accurately as possible and make accurate importance prediction.

[0050] However, in the actual operation of the power system, various real-time monitoring and intelligent peak shaving means have been taken to ensure the safety and stability of the power grid, and the real historical power flow data obtained often presents stable slight fluctuations or regular periodic tidal characteristics, which further makes the importance of different nodes and lines in the training set constructed based on the above-mentioned historical data have a strong correlation with the trend of change over time, and the change characteristics are single.

[0051] Using the training data set with the above characteristics to train the prediction model will result in the trained prediction model only having the ability to predict the importance of the power system running smoothly, and cannot accurately predict the power system importance when various sudden conditions cause the system power flow to fluctuate sharply, thereby greatly limiting the application scenarios and prediction accuracy of the above prediction model. Moreover, based on the characteristics of historical power flow data, the above defects cannot be solved simply by increasing the amount of training set data.

[0052] The training data set construction method provided by the present application is exactly to solve the above defects caused by the construction method of the training data set. Specifically, as shown in Figure 3 , the training data set construction method provided by the present application comprises the following steps:

[0053] S1: obtaining actual sampling data of a power system corresponding to a time sequence {t i}, wherein i∈[1,T];

[0054] S2: based on a randomly selected scenario mode, applying disturbance to the power system in a subsequence {t i} of the time sequence {t i′}, and calculating corresponding simulated sampling data, using the simulated sampling data to reconstruct the actual sampling data to obtain reconstructed sampling data of the power system;

[0055] S3: constructing the training data set based on the reconstructed sampling data.

[0056] The training data set construction method provided by the present application embodiment first obtains actual sampling data of the power system corresponding to each time point t i ,…t T} of a time sequence {t i}, then randomly selects a scenario mode, and applies disturbance to the power system in a subsequence {t i} of the time sequence {t i′} according to the selected scenario mode, to simulate the power change of each power generation / load unit, node, line and the change of power flow of the power system when various faults occur (the length of the disturbed subsequence {t i′} can be less than {t i}, or equal to {t i}, i.e. 1≤MIN(i′) and MAX(i′)≤T), and finally, based on the simulated sampling data after the disturbance, the training data set is generated using the reconstructed sampling data.

[0057] Specifically, in the embodiment of the present application, step S1 is used to obtain actual sampling data of a power system corresponding to a time sequence {t1,t2,…ti ,…t T actual sampling data of each node, line, power generation unit and load unit at each time point t

[0058] In some preferred embodiments of the present application, the actual sampling data comprises the actual sampling data of each node, line, power generation unit and load unit at each time point t i actual sampling values of corresponding active power and reactive power. In some specific embodiments, when there are M power generation units, N load units, K nodes and L lines in the power system, the active power and reactive power of each unit, node and line in the power system can be sampled at a fixed time interval (such as Δt = 15 minutes) for a continuous period of time (such as t T -t1= 250 hours), so as to obtain the active power and reactive power sampling values of each unit, node and line at multiple time points (such as 1000 time points), thereby obtaining the actual sampling data as shown in the following formula:

[0059]

[0060] wherein P and Q represent the active power and the reactive power respectively, and the subscripts represent the numbers of each unit, node and line. It should be noted that for each power generation / load unit, node and line in the power system, the above active power and reactive power are directional, such as for a power generation unit, its power is outgoing, for a load unit, its power is incoming, and for any line, the direction of its power is the direction of the power flow between the nodes connected thereto.

[0061] After obtaining the above actual sampling data, step S2 is configured to randomly select a scenario mode and perturb the power system accordingly, so as to simulate the changes of the active power and the reactive power of each unit, node and line under the sudden condition of occurrence of various faults, and to obtain the sampling data that can simulate various fault scenarios after superimposing the changes on the real sampling data.

[0062] In some preferred embodiments of the present application, the scenario mode comprises a power supply fault mode and a line fault mode.

[0063] Specifically, when the selected scenario mode comprises the power supply fault mode, at least one power generation unit is set to be faulty and its active power and reactive power are reduced compared with the actual sampling values, so as to simulate the situation that the power output of the power generation unit is reduced or even completely unable to output after the power generation unit is faulty.

[0064] Specifically, when the selected scenario mode comprises the line fault mode, at least one line is set to be faulty and its active power and reactive power are reduced to 0, so as to simulate the line breakage or short circuit fault caused by disasters such as lightning, storm, frost, snow and the like destroying a certain line.

[0065] In some embodiments of the present application, the result of the random selection can be a single power failure or a single line failure; in other embodiments of the present application, the result of the random selection can also be a simultaneous power failure or a simultaneous line failure; in still other embodiments of the present application, the above-mentioned power failure mode and line failure mode can also be selected in different sub-sequences to simulate the case of different failures occurring at different time periods. Those skilled in the art can flexibly select the above-mentioned scene mode without departing from the technical idea of the present application.

[0066] Specifically, in the embodiments of the present application, step S2 comprises the following steps:

[0067] S21: randomly selecting at least one scene mode;

[0068] S22: if the selected scene mode includes a power failure mode, further randomly selecting at least one power generation unit and setting its on-off state at each time point t i′ corresponding active power disturbance amount ΔP G (t i′ ) and reactive power disturbance amount ΔQ G (t i′ ), and

[0069] if the selected scene mode includes a line failure mode, further randomly selecting at least one line and setting its on-off state at each time point t i′ ;

[0070] S23: determining the corresponding simulation sampling data of time sequence {t i′} based on the ΔP G (t i′ ), ΔQ G (t i′ ) and on-off state change, wherein the simulation sampling data comprises the simulation sampling values of active power and reactive power of nodes, lines, power generation units and load units at each time point t i′ ;

[0071] S24: reconstructing the actual sampling data using the simulation sampling data to obtain the reconstructed sampling data of the power system.

[0072] Specifically, the calculation of the simulation sampling data in step S23 needs to satisfy the node power balance constraint. For example, when the scene mode includes a power failure mode, the reduced active power ΔP G (t i′ ) and reactive power ΔQ G (t i′ ) of the failed power generation unit at t i′ moment need to satisfy the node power balance constraint.The fault requires additional power output from other generating units to meet the node power balance limit until the fault is cleared after maintenance.

[0073] One optional regulation method follows the proximity principle, meaning that the nearest generating unit compensates for power generation without exceeding its output limit. In this regulation mode, the S nearest generating units can adjust their active and reactive power according to the following rules:

[0074]

[0075] Where the subscripts 1 to S represent the numbers of the power generation units adjacent to the faulty power generation unit, ΔP1 to ΔP s These are the additional active power added to each adjacent power generation unit, ΔQ1~ΔQ s These are the additional reactive power added to each adjacent power generation unit, and correspondingly, P K1 ~P KS and Q K1 ~Q KS These are its available active power and reactive power, respectively.

[0076] For example, when the scenario mode includes a line fault mode, at the time point t when the fault occurs... i′ The active and reactive power flowing through the faulty line will drop to 0, and the power will be transmitted by other lines, causing the active and reactive power of the surrounding lines and nodes to change accordingly to meet the node power balance limit, until the fault is eliminated after maintenance.

[0077] {t} is obtained through step S23. i′ After obtaining the corresponding simulated sampling data, in step S24, it can be used to analyze the time series sequence {t}. i The actual sampled data corresponding to} can be reconstructed. Specifically, the actual sampled data that corresponds to the subsequence {t} can be reconstructed. i′} at each time point t i′ The data at point t is replaced with simulated sampled data, while the data at other time points remain unchanged, finally resulting in the time series sequence {t}. i The corresponding reconstructed sampling data.

[0078] The reconstructed sampling data includes real data obtained by sampling the power system, as well as changes that occur when the power system is disturbed by simulating various fault conditions. In this way, the diversity of the training dataset can be significantly increased, thereby improving the training effect of the power system importance prediction model.

[0079] Specifically, step S3 further includes the following steps:

[0080] S31: establishing a directed weighted graph of the power system;

[0081] S32: obtaining time series {t i} of power flow data of the power system based on the reconstructed sampled data;

[0082] S33: calculating the importance of each node at each time point t i based on the power flow data and the directed weighted graph;

[0083] S34: calculating the importance of each line at each time point t i based on the power flow data and the directed weighted graph;

[0084] S35: generating the training data set, the training data set comprising the importance of each node and each line at each time point t i .

[0085] Preferably, the importance of each node at each time point t i is determined based on an EB-SALSA algorithm.

[0086] Preferably, the importance of each line at each time point t i is determined based on an EL-SALSA algorithm.

[0087] Specifically, the importance of each node can be calculated by the EB-SALSA algorithm and the importance of each line can be calculated by the EL-SALSA algorithm through the reconstructed sampled data as described above; in addition, the skilled person in the art can also select other preferred algorithms to calculate the importance of nodes and lines. The method of calculating the importance of nodes and lines based on the active power and reactive power data of each unit, node and line is known to the skilled person in the art and will not be described here.

[0088] Embodiment 1

[0089] In this embodiment, the data acquisition and the construction of the training data set are performed on the IEEE39 bus system shown in FIG. 1 using the method of constructing the training data set proposed in the present application. Figure 1

[0090] ​Firstly, the actual sampling data of the power system is acquired, wherein the sampling time interval is 15 minutes, 1500 time points are sampled in total, and the actual sampling values of 10 power generation units, 21 load units, 39 nodes and 46 lines at 1500 time points are acquired in total; then the power system is disturbed at the first to 141st time points, the simulation sampling values at the first to 141st time points are calculated, and the actual sampling data is reconstructed by using the simulation sampling values to obtain reconstructed sampling data; finally, the reconstructed sampling data is used to acquire a 39x1500 node importance data set and a 46x1500 line importance data set.

[0091] Figure 4 The change of the node importance of a specific node at the first to 141st time points is shown, and as a comparison, the change of the node importance obtained by using the conventional method is also shown in the figure.

[0092] Figure 5 The change of the line importance of a specific line at the first to 141st time points is shown, and as a comparison, the change of the line importance obtained by using the conventional method is also shown in the figure.

[0093] Embodiment 2

[0094] In this embodiment, the data acquisition and the construction of the training data set are performed on the IEEE118 bus system shown in the figure by using the construction method of the training data set proposed in the present application. Figure 2

[0095] Firstly, the actual sampling data of the power system is acquired, wherein the sampling time interval is 15 minutes, 1500 time points are sampled in total, and the actual sampling values of 10 power generation units, 21 load units, 39 nodes and 46 lines at 1500 time points are acquired in total; then the power system is disturbed at the first to 141st time points, the simulation sampling values at the first to 141st time points are calculated, and the actual sampling data is reconstructed by using the simulation sampling values to obtain reconstructed sampling data; finally, the reconstructed sampling data is used to acquire a 39x1500 node importance data set and a 46x1500 line importance data set.

[0096] Figure 6 The change of the node importance of a specific node at the first to 141st time points is shown, and as a comparison, the change of the node importance obtained by using the conventional method is also shown in the figure.

[0097] Figure 7 The change of the line importance of a specific line at the first to 141st time points is shown, and as a comparison, the change of the line importance obtained by using the conventional method is also shown in the figure. ​

[0098] By Figures 4 to 7 It can be seen that, compared with the node importance and line importance obtained by the real sampling data of the power system, the fluctuation of each node importance and line importance contained in the training data set constructed by the method provided in the application is more severe, and can effectively simulate the response of each node and line of the power system to various faults. The training effect of the power system importance prediction system using the training data set constructed by the method provided in the application can be more effective.

[0099] The specific embodiments of the application are described in detail above, and those skilled in the art can make some improvements and modifications to the application without departing from the principles of the application. These improvements and modifications also belong to the protection scope of the claims of the application.

Claims

1. A method for constructing a training data set, the training data set being used for training an importance prediction model of a power system, the power system comprising a plurality of nodes, a plurality of lines, a plurality of generation units and at least one load unit, characterized in that, The method comprises the following steps: S1: Obtain the actual sampling data of the power system corresponding to the time sequence {t i} where i ∈ [1, T]; S2: Scenario patterns based on random selection in {t i The subsequence {t} of} i′ A disturbance is applied to the power system and the corresponding simulated sampling data is calculated. The simulated sampling data is then used to reconstruct the actual sampling data to obtain the reconstructed sampling data of the power system. S3: constructing the training data set based on the reconstructed sampling data; Step S2 further comprises the following steps: S21: randomly selecting at least one scene mode; S22: If the selected scenario mode comprises a power failure mode, further randomly select at least one power generating unit and set its active power contribution ΔΡ(ί) and reactive power contribution ΔQ(ί) at each time point t i′ corresponding active power perturbation ΔΡ(ί) G (t i′ ) and reactive power perturbation ΔQ(ί) G (t i′ ), and, If the selected scenario mode comprises a line fault mode, at least one line is further randomly selected and its on-off state is set at each time point t i′ ; S23: determining the ΔP based on the ΔP G (t i′ ), ΔQ G (t i′ ) and the on-off state change determination time sequence {t i′} corresponding to the analog sampling data, the analog sampling data including the analog sampling values of the active power and the reactive power of the nodes, lines, power generation units and load units at each time point t i′ . S24: reconstructing the actual sampling data using the simulated sampling data to obtain the reconstructed sampling data of the power system; The power supply fault mode is specifically: The active power and the reactive power of the failed generating unit are reduced compared with the actual sampling value; The line fault mode is specifically: The active power and the reactive power of the failed line are reduced to 0; The calculation of the simulated sampling data satisfies the node power balance limit.

2. The method for constructing a training data set according to claim 1, characterized in that: The actual sampling data includes the nodes, lines, power generation units and load units at each time point t i The actual sampling values of the corresponding active power and reactive power.

3. The method for constructing a training data set according to claim 1, characterized in that: The scene mode comprises a power supply fault mode and a line fault mode.

4. The method of claim 1, wherein, Step S3 further comprises the following steps: S31: establishing a directed and weighted graph of the power system; S32: Obtain a time sequence {t i} corresponding to the power flow data of the power system based on the reconstructed sampling data S33: calculating, based on the power flow data and the directed and weighted graph, the importance of each node at each time point t i corresponding importance; S34: calculating, based on the power flow data and the directed weighted graph, the importance of each line at each time point t i corresponding importance; S35: generating the training data set, the training data set comprising each node and each line at each time point t i corresponding importance.

5. The method for constructing a training data set according to claim 4, characterized in that: Each node at each time point t i The corresponding importance is determined based on the EB-SALSA algorithm.

6. The method for constructing a training data set according to claim 4, characterized in that: Each line at each time point t i The corresponding importance is determined based on the EL-SALSA algorithm.

Images