A behavior simulation analysis method based on multi-source data fusion

Abstract

The present invention relates to the technical field of power system behavior analysis, and in particular to a behavior simulation analysis method based on multi-source data fusion. The method obtains a reference state correlation based on the correlation between the behavior state data of the power system and the behavior correlation data of different power correlation dimensions; obtains a reference behavior weight based on the reference state correlation and the fluctuation of the behavior correlation data in the power correlation dimension; and constructs a power behavior state evaluation model based on the reference behavior weight, so that the accuracy of the behavior simulation analysis of the power system based on the constructed power system behavior state evaluation model is higher.

Description

Technical Field

[0001] The present invention relates to the technical field of power system behavior analysis, and in particular to a behavior simulation analysis method based on multi-source data fusion. Background Art

[0002] Power system behavior simulation refers to the simulation and evaluation of power system behavior based on data related to power system behavior data, thereby analyzing the current behavior state or operating conditions of the power system based on the evaluation results, such as the degree of power system operation anomalies or the degree of power system stability. Traditional power system behavior simulation usually simulates power system behavior based on a single data source or a few data sources, such as the power system current, voltage, or load. However, with the expansion of power system scale and increasing complexity, traditional power behavior simulation methods cannot fully reflect the actual operating state of the power system due to the limited data sources, thus affecting the accuracy and predictive ability of the simulation analysis. Therefore, there is a need for an analysis method that can use multi-source power system related data to simulate power system behavior, so as to more comprehensively reflect the actual operating state of the power system.

[0003] Existing technologies typically use deep learning models to input comprehensive behavioral correlation data from multiple power-related dimensions into a trained deep learning model, outputting an evaluation value for the power system's behavioral state data. However, the behavioral correlation data in the power system is time-series and correlated data. Simply inputting the behavioral correlation data at each moment into a neural network model to evaluate the power system's behavioral state fails to consider the correlation between the behavioral correlation data and the behavioral state data, nor the time-series changes between the behavioral correlation data on a single power-related dimension. This approach has significant limitations, resulting in inaccurate evaluation values for the power system's behavioral state data. In other words, the existing method of inputting behavioral correlation data from multiple power-related dimensions into a deep learning model for power system behavior simulation analysis has low accuracy. Summary of the Invention

[0004] In order to solve the technical problem that the existing method of inputting behavioral correlation data of multiple power-related dimensions into a deep learning model to perform power system behavior simulation analysis has poor results, the purpose of the present invention is to provide a behavior simulation analysis method based on multi-source data fusion. The technical solution adopted is as follows:

[0005] The present invention proposes a behavior simulation analysis method based on multi-source data fusion, the method comprising:

[0006] At each sampling moment, the behavioral state data of the power system and the behavioral correlation data of each power-related dimension of the power system are obtained;

[0007] According to the linear numerical correlation between the behavior correlation data and the behavior state data of each power correlation dimension and the correlation of the data time series change, the reference state correlation of each power correlation dimension is obtained;

[0008] Obtaining a reference behavior weight for each power-related dimension based on the reference state association, a data value distribution deviation of the behavior-related data in each power-related dimension, and an overall data value;

[0009] A power system behavior state assessment model is constructed according to the reference behavior weights of various power-related dimensions; and a behavior simulation analysis of the power system is performed according to the power system behavior state assessment model.

[0010] Furthermore, the method for obtaining the reference state relevance includes:

[0011] The difference between each behavior state data at each sampling moment and the behavior state data at the next sampling moment is used as the behavior state change value at each sampling moment; in each power-related dimension, the difference between the behavior correlation data at each sampling moment and the behavior correlation data at the next sampling moment is used as the behavior correlation change value of each power-related dimension at each sampling moment;

[0012] Obtain an ascending sequence of behavior association change values for each power association dimension; arrange the behavior association change values in the ascending sequence in ascending order, and arrange equal behavior association change values in chronological order; use the index value of each behavior association change value in the ascending sequence as a reference level value for each behavior association change value in each power association dimension;

[0013] Obtaining an ascending sequence of behavior state change values; arranging the behavior state change values in the ascending sequence in ascending order, and arranging equal behavior change values in chronological order; using the index value of each behavior state change value in the ascending sequence as a reference level value for each behavior state change value;

[0014] In each power correlation dimension, a time series change correlation parameter calculation model is obtained based on the reference level value correlation distribution between the behavior correlation change value and the behavior state change value at each sampling moment; and a time series change correlation parameter of each power correlation dimension is obtained based on the time series change correlation parameter calculation model;

[0015] According to the time series correlation between the behavior state data at each sampling moment and the behavior correlation data of each power correlation dimension, the linear correlation parameter of each power correlation dimension is obtained;

[0016] The reference state correlation of each power correlation dimension is obtained according to the time series variation correlation parameter and the linear correlation parameter; the time series variation correlation parameter and the linear correlation parameter are both positively correlated with the reference state correlation.

[0017] Furthermore, the method for obtaining the reference behavior weight includes:

[0018] The mean of the behavior correlation data at all sampling moments corresponding to each power-related dimension is used as the reference correlation mean of each power-related dimension; the difference between the behavior correlation data of each power-related dimension at each sampling moment and the reference correlation mean is used as the behavior correlation deviation of each power-related dimension at each sampling moment; the mean of the behavior correlation deviations at all sampling moments corresponding to each power-related dimension is used as the volatility of the behavior correlation data of each power-related dimension;

[0019] According to the reference state correlation, the behavior correlation data volatility and the reference correlation mean, the reference behavior weight of each power correlation dimension is obtained; the reference state correlation and the behavior correlation data volatility are both negatively correlated with the reference behavior weight; the reference correlation mean is positively correlated with the correlation data volatility.

[0020] Furthermore, the method for constructing a power system behavior state assessment model based on the reference behavior weights of each power-related dimension includes:

[0021] The convolutional neural network is trained using the behavioral association data corresponding to all power-related dimensions at each sampling moment as input and the behavioral state data at each sampling moment as output.

[0022] Under each power-related dimension, data prediction is performed based on the behavior-related data at all sampling moments to obtain a predicted associated data value for each power-related dimension; the product of the predicted associated data value and the reference behavior weight is used as the weighted predicted data value for each power-related dimension;

[0023] The weighted predicted data values and the trained convolutional neural network are used as a power system behavior state assessment model.

[0024] Furthermore, the method for performing behavior simulation analysis on the power system according to the power system behavior state assessment model includes:

[0025] The weighted predicted data values of all power-related dimensions are input into the trained convolutional neural network, and the evaluation value of the behavioral status data of the power system is output.

[0026] Furthermore, the time series change associated parameter calculation model includes:

[0027]

[0028] Among them, Sp r is the time series variation correlation parameter of the rth power correlation dimension; n is the number of sampling moments; Δd i ′ is the reference level value of the behavior state change value at the i-th sampling moment; is the reference level value of the behavior correlation change value of the rth power correlation dimension at the i-th sampling moment; || is the absolute value symbol.

[0029] Furthermore, the method for obtaining the linear correlation parameter includes:

[0030] Arrange the behavior state data at all sampling moments in chronological order to obtain a behavior state data time series sequence; arrange the behavior association data of each power-related dimension at all sampling moments in chronological order to obtain a behavior association data time series sequence of each power-related dimension;

[0031] The Pearson correlation coefficient between the behavior state data time series sequence and the behavior association data time series sequence is used as the linear correlation parameter of each power association dimension.

[0032] Furthermore, the method for obtaining the reference state correlation of each power correlation dimension according to the time series variation correlation parameter and the linear correlation parameter includes:

[0033] The product of the time series variation correlation parameter and the linear correlation parameter is used as the reference state correlation of each power correlation dimension.

[0034] Furthermore, the method for obtaining the reference behavior weight of each power-related dimension according to the reference state correlation, the behavior-related data volatility and the reference correlation mean includes:

[0035] A positive correlation mapping value of the ratio of the product of the reference state relevance and the behavior relevance data volatility to the reference relevance mean is used as the reference behavior weight of each power relevance dimension.

[0036] Furthermore, the data prediction method adopts an autoregressive moving average model.

[0037] The present invention has the following beneficial effects:

[0038] The reason why the method of inputting behavioral correlation data of multiple power-related dimensions into the deep learning model for power system behavior simulation analysis has poor effect is that the behavioral correlation data at a single moment is input into the deep learning model without considering the correlation between the behavioral correlation data and the behavioral state data, as well as the temporal changes between the behavioral correlation data on a single power-related dimension. As a result, the evaluation value of the output behavioral state data has great limitations, which makes the accuracy of the power system behavior simulation analysis low. Therefore, if the accuracy of the power system behavior simulation analysis needs to be improved, it is necessary to combine the correlation between the behavioral correlation data of different dimensions and the temporal changes between the behavioral correlation data on a single power-related dimension before inputting the behavioral correlation data into the deep learning model. According to different correlations and temporal changes, the behavioral correlation data input into the deep learning model is affected, so as to perform more accurate power system behavior simulation.

[0039] The correlation between the behavior correlation data and the behavior state data of different power correlation dimensions is different. The greater the correlation of the behavior correlation data, the greater the impact of the size change of the behavior correlation data on the behavior state data. Therefore, the correlation calculation can be performed between the behavior correlation data and the behavior state data of each power correlation dimension, so that when the calculated reference state correlation is greater, the influence of the behavior correlation data of the corresponding power correlation dimension on the behavior state data is greater. First of all, considering that the behavior correlation data and the behavior state data in the power system are both time series data, the behavior correlation data and the behavior correlation data are correlated in time series; further considering that the behavior correlation data in the power system usually changes linearly or has a certain delay, the correlation analysis can be performed on the correlation of linear changes; therefore, the present invention obtains the reference state correlation of each power correlation dimension based on the linear numerical correlation between the behavior correlation data and the behavior state data of each power correlation dimension and the data time series change correlation, that is, the correlation between the behavior correlation data and the behavior state data of each power correlation dimension.

[0040] Furthermore, considering that there are many and complex power-related dimensions, the behavioral correlation data of multiple power-related dimensions will affect the behavioral state of the power system only when they are combined; therefore, changes in the behavioral state of the power system are often the joint influence of changes in the behavioral correlation data of multiple power-related dimensions; therefore, the more significant the change in the data of the power-related dimension or the more obvious the fluctuation, the greater the degree of change, and the greater the impact on the behavioral state of the power system; and because the reference state correlation represents the correlation between the behavioral correlation data of the power-related dimension and the behavioral state data; therefore, the greater the reference state correlation of the power-related dimension and the more significant the fluctuation of the corresponding behavioral correlation data, the greater the impact of the behavioral correlation data of the power-related dimension on the behavioral state; therefore, the present invention obtains the reference behavior weight of each power-related dimension based on the reference state correlation, the data value distribution deviation of the behavioral correlation data in each power-related dimension, and the overall data value, that is, the degree of influence of the behavioral correlation data of each power-related dimension on the behavioral state data of the power system; the greater the reference behavior weight, the greater the influence of the behavioral correlation data of the power-related dimension when constructing the power system behavioral state evaluation model. After further constructing a power system behavior state assessment model based on the reference behavior weights of each power-related dimension, a behavior simulation analysis of the power system is performed, so that the evaluation value of the power system behavior state data obtained is more accurate, that is, the accuracy of the behavior simulation analysis of the power system based on the power system behavior state assessment model is higher. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions and advantages of the embodiments of the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art descriptions. 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 paying any creative work.

[0042] Figure 1 A flowchart of a behavior simulation analysis method based on multi-source data fusion provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0043] To further illustrate the technical means and effectiveness of the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, describes in detail the specific implementation, structure, features, and effectiveness of a behavioral simulation analysis method based on multi-source data fusion proposed by the present invention. In the following description, different references to "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics of one or more embodiments may be combined in any suitable manner.

[0044] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0045] The following describes in detail a specific solution of a behavior simulation analysis method based on multi-source data fusion provided by the present invention with reference to the accompanying drawings.

[0046] See also Figure 1 , which shows a flow chart of a behavior simulation analysis method based on multi-source data fusion provided by one embodiment of the present invention, the method comprising:

[0047] Step S1: At each sampling moment, behavioral state data of the power system and behavioral correlation data of each power-related dimension of the power system are obtained.

[0048] An embodiment of the present invention aims to provide a behavioral simulation analysis method based on multi-source data fusion, which is used to analyze the correlation between the behavioral status data of the power system and the behavioral correlation data of each power-related dimension of the power system, as well as the temporal changes in the behavioral correlation data of each power-related dimension, to obtain a reference behavioral weight for each power-related dimension, and then construct a power behavior status evaluation model based on the reference behavioral weight, so that the accuracy of the behavioral simulation analysis of the power system based on the constructed power system behavior status evaluation model is higher.

[0049] Therefore, the embodiment of the present invention first obtains the behavioral state data of the power system and the behavioral association data of each power-related dimension of the power system at each sampling moment. In the embodiment of the present invention, the interval between sampling moments is set to 30 seconds, and the number of sampling moments is set to 50, which can be adjusted by the implementer according to the specific implementation environment. The behavioral state data of the power system uses the power system's operational abnormality data, while the power-related dimensions use voltage data, current data, electrical load data, generator temperature, and other data related to the power system's operational abnormality. These data are collected by installing corresponding sensors in the power system, such as voltage sensors to collect voltage data and current sensors to collect current data. This will not be further described here. The implementer can increase or decrease this data according to the specific implementation environment. This will not be further described here. As for the behavioral state data, the embodiment of the present invention uses the power system's operational abnormality as the behavioral state data, that is, the power system's operational abnormality level is collected at each moment. The specific method is to manually score the power system's operational abnormality level at each sampling moment and manually label the power system's operational abnormality level. In the embodiment of the present invention, the power system's operational abnormality level is generally divided into 10 numerical values, with the larger the numerical value of the operational abnormality level, the more abnormal the power system is. It should be noted that the degree of operational abnormality of the power system manually labeled here is the degree of operational abnormality during the labeling process in the subsequent convolutional neural network model training process, which is equivalent to collecting operational abnormality data of the power system while training the convolutional neural network model.

[0050] Step S2: Obtain the reference state correlation of each power correlation dimension based on the linear numerical correlation between the behavior correlation data and the behavior state data of each power correlation dimension and the data time series change correlation.

[0051] The reason why the method of inputting behavioral correlation data of multiple power-related dimensions into the deep learning model for power system behavior simulation analysis has poor effect is that the behavioral correlation data at a single moment is input into the deep learning model without considering the correlation between the behavioral correlation data and the behavioral state data, as well as the temporal changes between the behavioral correlation data on a single power-related dimension. As a result, the evaluation value of the output behavioral state data has great limitations, which makes the accuracy of the power system behavior simulation analysis low. Therefore, if the accuracy of the power system behavior simulation analysis needs to be improved, it is necessary to combine the correlation between the behavioral correlation data of different dimensions and the temporal changes between the behavioral correlation data on a single power-related dimension before inputting the behavioral correlation data into the deep learning model. According to different correlations and temporal changes, the behavioral correlation data input into the deep learning model is affected, so as to perform more accurate power system behavior simulation.

[0052] The correlation between the behavior correlation data and the behavior state data of different power correlation dimensions is different. The greater the correlation of the behavior correlation data, the greater the impact of the size change of the behavior correlation data on the behavior state data. Therefore, the correlation calculation can be performed between the behavior correlation data and the behavior state data of each power correlation dimension, so that when the calculated reference state correlation is greater, the influence of the behavior correlation data of the corresponding power correlation dimension on the behavior state data is greater. First of all, considering that the behavior correlation data and the behavior state data in the power system are both time series data, the behavior correlation data and the behavior correlation data are correlated in time series; further considering that the behavior correlation data in the power system usually changes linearly or has a certain delay, the correlation analysis can be performed on the correlation of linear changes; therefore, the embodiment of the present invention obtains the reference state correlation of each power correlation dimension based on the linear numerical correlation between the behavior correlation data and the behavior state data of each power correlation dimension and the data time series change correlation, that is, the correlation between the behavior correlation data and the behavior state data of each power correlation dimension.

[0053] Preferably, the method for obtaining the reference state association includes:

[0054] The difference between each behavioral state data at each sampling moment and the behavioral state data at the next sampling moment is used as the behavioral state change value at each sampling moment; in each power-related dimension, the difference between the behavioral association data at each sampling moment and the behavioral association data at the next sampling moment is used as the behavioral association change value of each power-related dimension at each sampling moment. When calculating the correlation between behavioral state data and behavioral association data, the correlation between the two time series can be directly obtained by using the calculation method of the Spearman rank correlation coefficient. However, considering that the trend and seasonality in the time series will affect the correlation calculation, the embodiment of the present invention eliminates the trend and seasonality in the time series by the difference method, making the data more stable, thereby improving the accuracy of the correlation calculated subsequently by the Spearman rank correlation coefficient; and for behavioral state data, the changes in behavioral state data are jointly affected by the changes in all behavioral association data. Therefore, the difference method can be used to analyze the degree of influence of the behavioral association data of each power-related dimension on the behavioral state data when the behavioral association data changes.

[0055] Obtain an ascending sequence of behavior correlation change values for each power correlation dimension; arrange the behavior correlation change values in the ascending sequence in ascending order, and arrange equal behavior correlation change values in chronological order; use the index value of each behavior correlation change value in the ascending sequence as the reference rank value for each behavior correlation change value in each power correlation dimension; obtain an ascending sequence of behavior state change values; arrange the behavior state change values in the ascending sequence in ascending order, and arrange equal behavior change values in chronological order; use the index value of each behavior state change value in the ascending sequence as the reference rank value for each behavior state change value. When calculating the Spearman rank correlation coefficient between two sets of data, it is usually necessary to assign a rank value to each data set. After arranging the data in an ascending order or a descending order, the index value of each data set after arrangement is used as the corresponding rank value for calculation. However, the traditional Spearman rank correlation coefficient will set the data that are equal after arrangement to the same size when assigning rank values; and the time series data in the embodiment of the present invention has a lag effect in time, and the rank values of the same size cannot be combined with this feature, so equal behavior state change values and equal behavior association change values are arranged in chronological order. For example, a group of data is arranged in order from small to large as (7, 11, 15, 15, 29), and the corresponding rank values in the traditional Spearman rank correlation coefficient are 1, 2, 3.5, 3.5, and 5. However, in the embodiment of the present invention, it is necessary to further analyze the sampling moments corresponding to the two values 15, and set a smaller rank value before the sampling moment, that is, divide the rank values corresponding to (7, 11, 15, 15, 29) into 1, 2, 3, 4, and 5, so that the same data corresponds to different rank values due to different times, so that the influence of time sequence is combined on the basis of linear numerical values.

[0056] In each power correlation dimension, a time series change correlation parameter calculation model is obtained based on the correlation distribution of the reference level values between the behavior correlation change values and the behavior state change values at each sampling moment; and the time series change correlation parameters of each power correlation dimension are obtained based on the time series change correlation parameter calculation model.

[0057] Preferably, the time series variation associated parameter calculation model includes:

[0058]

[0059] Among them, Sp r is the time series variation correlation parameter of the rth power correlation dimension; n is the number of sampling moments; Δd i ′ is the reference level value of the behavior state change value at the i-th sampling moment; is the reference level value of the behavioral correlation change value of the rth power correlation dimension at the i-th sampling moment; || is the absolute value symbol. The calculation model of the time series change correlation parameter is the same as the calculation method of the Spearman rank correlation parameter. The difference is that the data used is differential data and absolute value symbols, and the division method of the rank value is different; the purpose of the absolute value symbol here is to prevent the conflict with the sign of the subsequently calculated linear correlation parameter, resulting in a large error in the calculated reference state correlation. The calculation method of the Spearman rank correlation parameter is an existing technology well known to those skilled in the art, so the significance of the time series change correlation parameter calculation model will not be further elaborated here. In addition, it should be noted that the value range of the Spearman rank correlation parameter is -1 to 1, and the larger the absolute value, the stronger the correlation.

[0060] Further considering that the behavioral state data and the behavioral correlation data are linearly correlated in time, in order to make the calculated reference state correlation more accurate based on the time-series change correlation parameters, the embodiment of the present invention further obtains the linear correlation parameters of each power correlation dimension based on the time-series correlation between the behavioral state data at each sampling moment and the behavioral correlation data of each power correlation dimension.

[0061] Preferably, the method for obtaining the linear correlation parameter includes:

[0062] The behavioral state data at all sampling moments are arranged in chronological order to obtain a behavioral state data time series sequence; the behavioral state data at all sampling moments are arranged in chronological order to obtain a behavioral state data time series sequence for each power-related dimension; changes in the behavioral state data of the power system result in changes in the linear correlation parameters of all power-related dimensions; therefore, the behavioral state data time series sequence and the behavioral correlation data time series sequence are linearly correlated. The Pearson correlation coefficient can intuitively reflect the degree of linear relationship between the two sequences. Therefore, the Pearson correlation coefficient between the behavioral state data time series sequence and the behavioral correlation data time series sequence is used as the linear correlation parameter for each power-related dimension. The Pearson correlation coefficient also ranges from -1 to 1. The larger the corresponding absolute value, the stronger the correlation. A value closer to -1 indicates a stronger negative correlation, and a value closer to 1 indicates a stronger positive correlation. It should be noted that the Pearson correlation coefficient is a prior art well known to those skilled in the art and will not be further defined or elaborated upon here.

[0063] Further, based on the time series variation correlation parameter and the linear correlation parameter, the reference state correlation of each power correlation dimension is obtained; both the time series variation correlation parameter and the linear correlation parameter are positively correlated with the reference state correlation. Preferably, the method for obtaining the reference state correlation of each power correlation dimension based on the time series variation correlation parameter and the linear correlation parameter includes:

[0064] Since both the time series change correlation parameter and the linear correlation parameter can characterize the correlation between the behavior correlation coefficient and the behavior state coefficient, the product of the time series change correlation parameter and the linear correlation parameter is further used as the reference state correlation of each power correlation dimension. Since the time series change correlation parameter is obtained by the absolute value, after the product combination, when the linear correlation parameter is a positive number, the calculated reference state correlation is greater than or equal to 0, which means that the behavior correlation coefficient of the power correlation dimension is positively correlated with the behavior state coefficient or has no correlation relationship; when the linear correlation parameter is a negative number, the calculated reference state correlation is less than or equal to 0, which means that the behavior correlation coefficient of the power correlation dimension is negatively correlated with the behavior state coefficient or has no correlation relationship. And when the reference behavior weight is affected by the reference state correlation, the suppression or enhancement of the data can be achieved through positive and negative correlation, which is more in line with the purpose of adjusting the weight in the embodiment of the present invention. Therefore, the reference state correlation obtained by the product is more accurate.

[0065] In the embodiment of the present invention, each power correlation dimension is sequentially used as the kth power correlation dimension, and the method for obtaining the reference state correlation of the kth power correlation dimension is expressed in the formula as follows:

[0066] H k =T k ×X k

[0067] Among them, H k is the reference state correlation of the kth power correlation dimension; T k is the time series variation correlation parameter of the kth power correlation dimension; X k is the linear correlation parameter of the kth power correlation dimension.

[0068] Step S3: Obtain a reference behavior weight for each power-related dimension based on the reference state relevance, the data value distribution deviation of the behavior-related data in each power-related dimension, and the overall data value.

[0069] Furthermore, considering that there are many and complex power-related dimensions, the behavioral correlation data of multiple power-related dimensions will affect the behavioral state of the power system only when they are combined; therefore, changes in the behavioral state of the power system are often the joint influence of changes in the behavioral correlation data of multiple power-related dimensions; therefore, the more significant the change in the data of the power-related dimension or the more obvious the fluctuation, the greater the degree of change, and the greater the impact on the behavioral state of the power system; and because the reference state correlation represents the correlation between the behavioral correlation data of the power-related dimension and the behavioral state data; therefore, the greater the reference state correlation of the power-related dimension and the more significant the fluctuation of the corresponding behavioral correlation data, the greater the impact of the behavioral correlation data of the power-related dimension on the behavioral state; therefore, the embodiment of the present invention obtains the reference behavior weight of each power-related dimension based on the reference state correlation, the data value distribution deviation of the behavioral correlation data in each power-related dimension, and the overall data value, that is, the degree of influence of the behavioral correlation data of each power-related dimension on the behavioral state data of the power system; the greater the reference behavior weight, the greater the influence of the behavioral correlation data of the power-related dimension when constructing the power system behavioral state evaluation model.

[0070] Preferably, the method for obtaining the reference behavior weight includes:

[0071] The mean of the behavioral correlation data for all sampling moments corresponding to each power-related dimension is used as the reference correlation mean for each power-related dimension. The difference between the behavioral correlation data for each power-related dimension at each sampling moment and the reference correlation mean is used as the behavioral correlation deviation for each power-related dimension at each sampling moment. The mean of the behavioral correlation deviations for all sampling moments corresponding to each power-related dimension is used as the volatility of the behavioral correlation data for each power-related dimension. Because greater variability in the behavioral correlation data for a power-related dimension has a greater impact on the behavioral state of the power system, the corresponding reference behavior weight should be greater. Therefore, the greater the overall behavioral correlation deviation for the power-related dimension, that is, the greater the volatility of the behavioral correlation data, the greater the reference behavior weight.

[0072] Since the reference state relevance can affect the reference behavior weight, and the greater the reference state relevance, the greater the influence of the behavior-related data of the corresponding power-related dimension on the behavior state data, and the corresponding reference behavior weight is also greater. The mean of the behavior-related data is used to unify the dimension of the volatility of the behavior-related data, making the calculated reference behavior weight more robust. Therefore, the embodiment of the present invention obtains the reference behavior weight for each power-related dimension based on the reference state relevance, the volatility of the behavior-related data, and the reference correlation mean; the reference state relevance and the volatility of the behavior-related data are both negatively correlated with the reference behavior weight; and the reference correlation mean is positively correlated with the volatility of the correlation data.

[0073] Preferably, the method for obtaining the reference behavior weight of each power-related dimension according to the reference state correlation, the behavior-related data volatility and the reference correlation mean includes:

[0074] The positive correlation mapping value of the ratio of the product of the reference state correlation and the volatility of the behavior correlation data to the reference correlation mean is used as the reference behavior weight of each power correlation dimension. First, the greater the volatility of the behavior correlation data, the greater the influence of the corresponding power correlation dimension on the behavior correlation data and the behavior state data; and when the reference state correlation does not consider the sign, that is, when the positive or negative correlation is not considered, the greater the corresponding absolute value, the greater the correlation between the behavior state data and the behavior correlation data, and the greater the product between the absolute value of the reference state correlation and the behavior correlation data, the greater the influence of the behavior correlation data of the corresponding power correlation dimension on the behavior state data;

[0075] Further analysis was conducted on the positively correlated power-related dimensions and the negatively correlated power-related dimensions. For the positively correlated power-related dimensions, due to the positive correlation, the greater the increase in the corresponding behavior-related data, the greater the corresponding impact. The reference state correlation corresponding to the positively correlated power-related dimensions is positive, so the larger the reference behavior weight obtained by direct multiplication, the greater the increase in the weighted behavior-related data, and the greater the corresponding impact. For the negatively correlated power-related dimensions, due to the negative correlation, the greater the decrease in the corresponding behavior-related data, the greater the corresponding impact. The reference state correlation corresponding to the negatively correlated power-related dimensions is negative, so the smaller the reference behavior weight obtained by direct multiplication, that is, the larger the absolute value, the greater the decrease in the weighted behavior-related data, and the smaller the corresponding impact.

[0076] In the embodiment of the present invention, the method for obtaining the reference behavior weight of the kth power-related dimension is expressed in the formula:

[0077]

[0078] Among them, w k is the reference behavior weight of the kth power-related dimension; H k is the reference state correlation of the kth power correlation dimension; N k is the number of behavior-related data in the kth power-related dimension, that is, the number of sampling moments; g k,i is the behavior correlation data at the i-th sampling moment in the k-th power correlation dimension; is the mean of the behavior correlation data at all sampling moments in the k-th power correlation dimension. is the behavioral correlation deviation at the i-th sampling moment in the k-th power correlation dimension; is the volatility of the behavioral correlation data of the kth power correlation dimension. k When it is a negative value, The overall value is negative. After further adding the value 1, the reference behavior weight obtained is less than 1. After weighting the behavior correlation data of the power correlation dimension by the reference behavior weight, the larger the corresponding absolute value, the more the behavior correlation data is reduced. In other words, the more the behavior correlation data corresponding to the power correlation dimension that meets the negative correlation is reduced, the greater the corresponding impact degree is. Correspondingly, H k When it is positive, The overall value is positive. After further adding the value 1, the reference behavior weight obtained is greater than 1. After weighting the behavior correlation data of the power correlation dimension by the reference behavior weight, the larger the corresponding absolute value, the more the behavior correlation data increases. That is, the more the behavior correlation data corresponding to the power correlation dimension that meets the positive correlation increases, the greater the corresponding impact.

[0079] Step S4: constructing a power system behavior state assessment model according to the reference behavior weights of each power-related dimension; and performing behavior simulation analysis on the power system according to the power system behavior state assessment model.

[0080] Furthermore, a power system behavior state evaluation model is constructed based on the reference behavior weights of each power-related dimension. Preferably, the method for constructing a power system behavior state evaluation model based on the reference behavior weights of each power-related dimension includes:

[0081] The convolutional neural network is trained with the behavioral correlation data corresponding to all power-related dimensions at each sampling moment as input and the behavioral state data at each sampling moment as output; in each power-related dimension, data prediction is performed based on the behavioral correlation data at all sampling moments to obtain the predicted correlation data value of each power-related dimension; preferably, the data prediction method adopts an autoregressive moving average model. First, the power system behavior state is obtained by evaluating the behavioral correlation data of each power-related dimension, which itself is unpredictable, but the behavioral correlation data of each power-related dimension is predictable. Therefore, after predicting the behavioral correlation data of each power-related dimension, the power system behavior state can be evaluated and predicted based on the predicted values, so that the predicted power system behavior state is prepared in advance. It should be noted that the autoregressive moving average model and the convolutional neural network are existing technologies well known to those skilled in the art. Those skilled in the art can replace the autoregressive moving average model with other data prediction methods, or replace the convolutional neural network with other deep learning models. No further limitation or elaboration is made here. Furthermore, considering that the degree of influence between data of different power-related dimensions and behavioral status data is different, that is, the reference behavior weights are different, the predicted correlation data obtained by predicting each power-related dimension is further weighted by the reference behavior weight, that is, the product of the predicted correlation data value and the reference behavior weight is used as the weighted predicted data value of each power-related dimension; thereby, the weighted predicted data value combines the correlation between behavioral correlation data of different dimensions and the temporal changes between behavioral correlation data on a single power-related dimension to perform behavioral status data analysis.

[0082] Furthermore, the weighted predicted data values and the trained convolutional neural network are used as a power system behavior state assessment model; that is, the data of each power-related dimension is input into the trained convolutional neural network, and the embodiment of the present invention performs a behavior simulation analysis on the power system according to the power system behavior state assessment model. Preferably, the method for performing a behavior simulation analysis on the power system according to the power system behavior state assessment model includes:

[0083] The embodiment of the present invention inputs the weighted predicted data values of all power-related dimensions into a trained convolutional neural network, and outputs the evaluation value of the behavioral status data of the power system, that is, the predicted value of the degree of operational abnormality of the power system, so that the predicted value of the degree of operational abnormality can be responded to in advance. When the predicted value of the degree of operational abnormality is greater than 8, the embodiment of the present invention issues an abnormality warning to notify personnel to conduct an abnormality inspection on the power system, and the implementer can also set the threshold value of the abnormality warning by himself.

[0084] To sum up, the present invention obtains the reference state correlation based on the correlation between the behavioral state data of the power system and the behavioral correlation data of different power correlation dimensions; obtains the reference behavior weight based on the reference state correlation and the fluctuation of the behavioral correlation data in the power correlation dimension; and constructs the power behavior state evaluation model according to the reference behavior weight, so that the accuracy of the behavioral simulation analysis of the power system based on the constructed power system behavior state evaluation model is higher.

[0085] It should be noted that the order in which the embodiments of the present invention are described above is for illustrative purposes only and does not necessarily represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or sequential order shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0086] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments.

Claims

1. A power system behavior simulation analysis method based on multi-source data fusion, characterized in that: The method comprises: At each sampling moment, behavioral state data of the power system and behavioral association data of each power-related dimension of the power system are obtained; wherein the behavioral state data of the power system is data on the degree of abnormal operation of the power system, and the behavioral data of each power-related dimension is data associated with the degree of abnormal operation of the power system, including voltage data, current data, electric load data, and generator temperature; According to the linear numerical correlation between the behavior correlation data and the behavior state data of each power correlation dimension and the correlation of the data time series change, the reference state correlation of each power correlation dimension is obtained; Obtaining a reference behavior weight for each power-related dimension based on the reference state association, a data value distribution deviation of the behavior-related data in each power-related dimension, and an overall data value; A power system behavior state assessment model is constructed according to the reference behavior weights of various power-related dimensions; and a behavior simulation analysis of the power system is performed according to the power system behavior state assessment model.

2. The method for simulating and analyzing power system behavior based on multi-source data fusion according to claim 1, characterized in that: The method for obtaining the reference state association includes: The difference between each behavior state data at each sampling moment and the behavior state data at the next sampling moment is used as the behavior state change value at each sampling moment; in each power-related dimension, the difference between the behavior correlation data at each sampling moment and the behavior correlation data at the next sampling moment is used as the behavior correlation change value of each power-related dimension at each sampling moment; Obtain an ascending sequence of behavior association change values for each power association dimension; arrange the behavior association change values in the ascending sequence in ascending order, and arrange equal behavior association change values in chronological order; use the index value of each behavior association change value in the ascending sequence as a reference level value for each behavior association change value in each power association dimension; Obtaining an ascending sequence of behavior state change values; arranging the behavior state change values in the ascending sequence in ascending order, and arranging equal behavior change values in chronological order; using the index value of each behavior state change value in the ascending sequence as a reference level value for each behavior state change value; In each power correlation dimension, a time series change correlation parameter calculation model is obtained based on the reference level value correlation distribution between the behavior correlation change value and the behavior state change value at each sampling moment; and a time series change correlation parameter of each power correlation dimension is obtained based on the time series change correlation parameter calculation model; According to the time series correlation between the behavior state data at each sampling moment and the behavior correlation data of each power correlation dimension, the linear correlation parameter of each power correlation dimension is obtained; The reference state correlation of each power correlation dimension is obtained according to the time series variation correlation parameter and the linear correlation parameter; the time series variation correlation parameter and the linear correlation parameter are both positively correlated with the reference state correlation.

3. The power system behavior simulation analysis method based on multi-source data fusion according to claim 1 is characterized in that: The method for obtaining the reference behavior weight includes: The mean of the behavior correlation data at all sampling moments corresponding to each power-related dimension is used as the reference correlation mean of each power-related dimension; the difference between the behavior correlation data of each power-related dimension at each sampling moment and the reference correlation mean is used as the behavior correlation deviation of each power-related dimension at each sampling moment; the mean of the behavior correlation deviations at all sampling moments corresponding to each power-related dimension is used as the volatility of the behavior correlation data of each power-related dimension; According to the reference state correlation, the behavior correlation data volatility and the reference correlation mean, the reference behavior weight of each power correlation dimension is obtained; the reference state correlation and the behavior correlation data volatility are both negatively correlated with the reference behavior weight; the reference correlation mean is positively correlated with the correlation data volatility.

4. The method for simulating and analyzing power system behavior based on multi-source data fusion according to claim 1, characterized in that: The method for constructing a power system behavior state assessment model based on reference behavior weights of various power-related dimensions includes: The convolutional neural network is trained using the behavioral association data corresponding to all power-related dimensions at each sampling moment as input and the behavioral state data at each sampling moment as output. Under each power-related dimension, data prediction is performed based on the behavior-related data at all sampling moments to obtain a predicted associated data value for each power-related dimension; the product of the predicted associated data value and the reference behavior weight is used as the weighted predicted data value for each power-related dimension; The weighted predicted data values and the trained convolutional neural network are used as a power system behavior state assessment model.

5. The method for simulating and analyzing power system behavior based on multi-source data fusion according to claim 4 is characterized in that: The method for performing behavior simulation analysis on the power system according to the power system behavior state assessment model comprises: The weighted predicted data values of all power-related dimensions are input into the trained convolutional neural network, and the evaluation value of the behavioral status data of the power system is output.

6. The method for simulating and analyzing power system behavior based on multi-source data fusion according to claim 2, characterized in that: The temporal variation associated parameter calculation model includes: Among them, Sp r is the time series variation correlation parameter of the rth power correlation dimension; n is the number of sampling moments; Δd i ′ is the reference level value of the behavior state change value at the i-th sampling moment; is the reference level value of the behavior correlation change value of the rth power correlation dimension at the i-th sampling moment; || is the absolute value symbol.

7. The method for simulating and analyzing power system behavior based on multi-source data fusion according to claim 2, characterized in that: The method for obtaining the linear correlation parameter includes: Arrange the behavior state data at all sampling moments in chronological order to obtain a behavior state data time series sequence; arrange the behavior association data of each power-related dimension at all sampling moments in chronological order to obtain a behavior association data time series sequence of each power-related dimension; The Pearson correlation coefficient between the behavior state data time series sequence and the behavior association data time series sequence is used as the linear correlation parameter of each power association dimension.

8. The method for simulating and analyzing power system behavior based on multi-source data fusion according to claim 2, characterized in that: The method for obtaining the reference state correlation of each power correlation dimension according to the time series variation correlation parameter and the linear correlation parameter includes: The product of the time series variation correlation parameter and the linear correlation parameter is used as the reference state correlation of each power correlation dimension.

9. The method for simulating and analyzing power system behavior based on multi-source data fusion according to claim 3 is characterized in that: The method for obtaining the reference behavior weight of each power-related dimension according to the reference state correlation, the behavior-related data volatility, and the reference correlation mean includes: A positive correlation mapping value of the ratio of the product of the reference state relevance and the behavior relevance data volatility to the reference relevance mean is used as the reference behavior weight of each power relevance dimension.

10. The method for simulating and analyzing power system behavior based on multi-source data fusion according to claim 4, characterized in that: The data prediction method adopts an autoregressive moving average model.

Images