Power system frequency response lowest point prediction method, device, equipment and medium

Abstract

The invention discloses a power system frequency response lowest point prediction method and device, equipment and a storage medium, which are used for improving the prediction accuracy of the power system frequency response lowest point. The method comprises the following steps: performing transient simulation on a power system network, collecting a frequency data set, and extracting a frequency response lowest point data set in the frequency data set; the frequency data set comprises a plurality of pieces of time sequence data; converting the time sequence data into distance data to obtain a distance data set; performing regression learning on the distance data set and the frequency response lowest point data set, and establishing a lowest point prediction model; when it is detected that the power system network has a fault, obtaining actually measured frequency sample time sequence data of the power system network; converting the actually measured frequency sample time sequence data into target distance data; and inputting the target distance data into the lowest point prediction model to obtain a frequency response lowest point prediction result of the actually measured frequency sample data.

Description

Technical Field

[0001] The present invention relates to the technical field of frequency response prediction, and in particular to a method, device, equipment and medium for predicting the lowest point of frequency response of an electric power system. Background Art

[0002] With the development trend of new power systems, the proportion of renewable energy power is gradually increasing, but it also brings about frequency stability problems. As the proportion of thermal power units decreases, the system's rotational inertia decreases, and the anti-disturbance ability decreases accordingly. When the system has a large power disturbance causing frequency deviation, it will not be able to fully assume the function of supporting the system frequency. At the same time, the relatively high proportion of renewable energy units does not have the rotational inertia and frequency regulation capabilities of thermal power units due to the power electronic interface they use, making the frequency stability problem of the power system more prominent.

[0003] Therefore, if the frequency response characteristics of the system can be accurately predicted and the frequency instability risk can be assessed at the early stage of the disturbance, and resources can be mobilized in time for frequency support, the frequency stability of the system can be effectively guaranteed. At present, the common methods for studying the frequency problems of power systems under transient conditions can be divided into model-driven methods and data-driven methods. Among them, the model-driven method includes time domain simulation method and equivalent model method. The time domain simulation method is based on numerical analysis to carry out detailed modeling of various components in the system. Although this method is relatively accurate, it has a large amount of calculation and a long calculation time, and is not suitable for online prediction of frequency response characteristics. The equivalent model method is based on mathematical analysis and can be used for online calculation. The commonly used methods are divided into average system frequency model (Average System Frequency, ASF) and system frequency response model (System Frequency Response, SFR). The parameters of the equivalent model method are convenient for the calculation and formulation of frequency control strategies in the later stage, but due to the changes in the output of new energy in the power system, the model parameters will also change accordingly. If fixed model parameters are still used, the frequency response characteristics of the system can no longer be accurately reflected, which will affect the accuracy of frequency control strategy formulation.

[0004] With the development of science and technology, the hardware performance of computers has been improved, and analysis and prediction methods based on artificial intelligence have also been promoted in the power industry. In recent years, a large number of high-precision measurement devices such as synchronous phasor measurement units (PMUs) have been installed in my country's power systems. The past operation data stored in the PMU master station can help provide data support for training artificial intelligence algorithms. At the same time, its high-precision, high-frequency sampling and extremely fast data transmission rate can also help to achieve real-time perception of the operating status of the power grid.

[0005] This method is generally based on the framework of a machine learning model, using historical disturbance samples to train the model, using system operation data as input, and outputting system frequency response characteristics. Machine learning provides a reliable method for understanding the frequency situation of power systems and analyzing system frequency responses, but when using this method, the input data is of many types and dimensions, and the prediction accuracy depends on the selection of machine learning models. Summary of the invention

[0006] The present invention provides a method, device, equipment and storage medium for predicting the lowest point of frequency response of an electric power system, which are used to improve the accuracy of predicting the lowest point of frequency response of an electric power system.

[0007] The present invention provides a method for predicting the lowest point of frequency response of a power system, comprising:

[0008] Performing transient simulation on the power system network, collecting frequency data sets, and extracting frequency response minimum point data sets from the frequency data sets; the frequency data sets include a plurality of time series data;

[0009] Converting the time series data into distance data to obtain a distance data set;

[0010] Performing regression learning on the distance data set and the frequency response lowest point data set to establish a lowest point prediction model;

[0011] When a fault is detected in the power system network, obtaining time series data of measured frequency samples of the power system network;

[0012] Convert measured frequency sample time series data into target distance data;

[0013] The target distance data is input into the lowest point prediction model to obtain the frequency response lowest point prediction result of the measured frequency sample data.

[0014] Optionally, the step of converting the measured frequency sample time series data into target distance data includes:

[0015] Extracting the optimal time series feature Shapelets from the measured frequency sample time series data according to the learning time series algorithm LTS;

[0016] The multidimensional distance data of the time series data is extracted according to the optimal time series feature Shapelets as the target distance data corresponding to the time series data.

[0017] Optionally, the step of performing regression learning on the distance data set and the frequency response lowest point data set to establish a lowest point prediction model includes:

[0018] Support vector regression (SVR) is used to perform regression learning on the distance data set and the frequency response lowest point data set to establish a lowest point prediction model.

[0019] Optionally, the step of extracting optimal time series feature Shapelets from the measured frequency sample time series data according to the learning time series algorithm LTS includes:

[0020] Obtaining initial time series characteristics of the measured frequency sample time series data;

[0021] The initial time series features are iteratively optimized by minimizing the regression loss function to generate the optimal time series feature Shapelets of the measured frequency sample time series data.

[0022] The present invention also provides a device for predicting the lowest point of frequency response of a power system, comprising:

[0023] A transient simulation module is used to perform transient simulation on the power system network, collect frequency data sets, and extract frequency response minimum point data sets from the frequency data sets; the frequency data sets include a number of time series data;

[0024] A distance data set acquisition module, used for converting the time series data into distance data to obtain a distance data set;

[0025] A lowest point prediction model building module is used to perform regression learning on the distance data set and the frequency response lowest point data set to build a lowest point prediction model;

[0026] A measured frequency sample time series data acquisition module is used to acquire the measured frequency sample time series data of the power system network when a fault is detected in the power system network;

[0027] A target distance data conversion module is used to convert the measured frequency sample time series data into target distance data;

[0028] The frequency response lowest point prediction module is used to input the target distance data into the lowest point prediction model to obtain the frequency response lowest point prediction result of the measured frequency sample data.

[0029] Optionally, the target distance data conversion module includes:

[0030] The optimal time series feature Shapelets extraction submodule is used to extract the optimal time series feature Shapelets in the measured frequency sample time series data according to the learning time series algorithm LTS;

[0031] The target distance data conversion submodule is used to extract the multidimensional distance data of the time series data according to the optimal time series feature Shapelets as the target distance data corresponding to the time series data.

[0032] Optionally, the lowest point prediction model building module includes:

[0033] The lowest point prediction model establishment submodule is used to use support vector regression SVR to perform regression learning on the distance data set and the frequency response lowest point data set to establish a lowest point prediction model.

[0034] Optionally, the optimal time series feature Shapelets extraction submodule includes:

[0035] An initial time series feature acquisition unit, used to acquire the initial time series features of the measured frequency sample time series data;

[0036] The optimal time series feature Shapelets generating unit is used to iteratively optimize the initial time series features by minimizing the regression loss function to generate the optimal time series feature Shapelets of the measured frequency sample time series data.

[0037] The present invention also provides an electronic device, the device comprising a processor and a memory:

[0038] The memory is used to store program code and transmit the program code to the processor;

[0039] The processor is used to execute the method for predicting the lowest point of frequency response of a power system as described in any one of the above items according to the instructions in the program code.

[0040] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium is used to store program code, and the program code is used to execute the method for predicting the lowest point of frequency response of a power system as described in any one of the above items.

[0041] It can be seen from the above technical scheme that the present invention has the following advantages: the present invention discloses a method for predicting the lowest point of the frequency response of a power system, and specifically discloses: performing transient simulation on the power system network, collecting frequency data sets; and extracting the frequency response lowest point data set in the frequency data set; the frequency data set contains a number of time series data; converting the time series data into distance data to obtain a distance data set; performing regression learning on the distance data set and the frequency response lowest point data set to establish a lowest point prediction model; when a fault is detected in the power system network, obtaining the measured frequency sample time series data of the power system network; converting the measured frequency sample time series data into target distance data; inputting the target distance data into the lowest point prediction model to obtain the frequency response lowest point prediction result of the measured frequency sample data. The present invention retains the time series characteristics of the time series data by converting the time series data into distance data, thereby improving the prediction accuracy of the lowest point of the frequency response of the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. 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 creative labor.

[0043] Figure 1 A flowchart of a method for predicting the lowest point of frequency response of a power system provided by an embodiment of the present invention;

[0044] Figure 2 is the candidate subsequence S k With a time series T of length Q i Schematic diagram of the distance calculation process;

[0045] Figure 3 This is a flowchart of the algorithm for learning time series;

[0046] Figure 4 Schematic diagram of support vector regression;

[0047] Figure 5 A prediction flow chart of the lowest point of frequency response of a power system provided by an embodiment of the present invention;

[0048] Figure 6 A structural block diagram of a device for predicting the lowest point of frequency response of a power system provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0049] The embodiments of the present invention provide a method, device, equipment and storage medium for predicting the lowest point of frequency response of a power system, which are used to improve the accuracy of predicting the lowest point of frequency response of a power system.

[0050] In order to make the purpose, features and advantages of the present invention more obvious and easy to understand, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described below are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0051] See also Figure 1 , Figure 1 A flowchart of a method for predicting the lowest point of frequency response of a power system provided by an embodiment of the present invention.

[0052] The present invention provides a method for predicting the lowest point of a power system frequency response, which may specifically include the following steps:

[0053] Step 101, perform transient simulation on the power system network, collect frequency data sets; and extract the frequency response lowest point data set in the frequency data sets; the frequency data sets include a plurality of time series data;

[0054] Step 102, converting the time series data into distance data to obtain a distance data set;

[0055] In the embodiment of the present invention, the time series data may be converted into distance data through Shapelet transformation.

[0056] Shapelet is a subsequence-based time series classification method. The core idea of ​​this method is to identify a time series as a certain category based on discriminative subsequences. These discriminative subsequences are defined as Shapelets, which can be regarded as a shape feature of time series data. They can capture the key differences in the time series and classify them based on this.

[0057] The original Shapelet method aims to embed the Shapelet discovery mechanism in the process of decision tree construction. In essence, the Shapelet method is defined as a decision tree classifier, which is divided into two phases: training and testing. The training phase involves the establishment of a decision tree and identifying the most discriminating subsequences, i.e., Shapelets, at each node of the tree. In the testing phase, a new time series is classified based on the decision tree and its Shapelet. The following is a detailed description of the method for finding Shapelets at each node in the training phase:

[0058] (1) Generate a candidate subsequence set: Assume that the data set consists of I samples and each sequence contains Q ordered values. Then the time series data set is defined as , set the maximum and minimum lengths of the candidate subsequences to MAXLEN and MINLEN respectively, use the sliding window method to obtain all subsequences with lengths between MINLEN and MAXLEN and add them to the subsequence candidate set. For example, on a time series t, slide the sliding window of length L in the time axis direction with a step of 1, and you can get all subsequences of length L in the time series. On this basis, use this method to traverse all time series in the data set and get all subsequences in turn.

[0059] (2) Calculate the distance and take the subsequence with the maximum information gain as the Shapelet: First, process all candidate subsequences and calculate their distances from each time series in the data set. Since the length of the candidate subsequence is inconsistent with that of the time series sample, it cannot be calculated directly. Therefore, if we want to calculate a candidate subsequence S with a length of L k With a time series T of length Q i If the distance between k With T i All with S k The distances of subsequences of equal length are then selected as S k With T i The specific implementation steps are as follows: Slide the window with a length of L at a step size of 1 at T i Slide up to generate P subsequences, where J:=Q-L+1. i The subsequence starting from the jth time step and with a length of L is denoted as T i, j: j+L-1 , so we have:

[0060] (1)

[0061] Among them, S k With T i The distance between can be calculated as follows:

[0062] (2)

[0063] The distance calculation process is as follows Figure 2 As shown, the essence of calculating the distance between the two is to find the subsequence with the shortest distance in the time series corresponding to the candidate subsequence, and use this as the distance between the two.

[0064] After obtaining the distance dataset of the candidate subsequences and the time series, an optimal split point (OSP) needs to be determined. The main function of this split point is to split the dataset based on the distance value, where sequences greater than the OSP are divided into the right subtree, and sequences less than the OSP are divided into the left subtree. The information gain is calculated based on the information entropy before and after the split. If the proportion of class C samples in the dataset D of the current node is , then the information entropy calculation formula is as follows:

[0065] (3)

[0066] According to the optimal partition node, the sample is divided into left and right subtrees. The left subtree data set is recorded as D1 and the right subtree data set is recorded as D2. The information gain is calculated as follows:

[0067] (4)

[0068] The partition point with the largest information gain is called the optimal partition point, but in principle, there are countless values ​​of the partition point, which cannot be solved by traversal. Therefore, the binary search method is used, that is, the distance data is sorted by size, the mean of the adjacent data is used as the partition point and the information gain is calculated. The partition point with the largest information gain is the optimal partition point. This method calculates the information gain of all candidate subsequences, and retains the subsequence with the largest information gain and the optimal partition point at each node, and divides the distance data into left and right subtrees based on this. During testing, the distance between the input data and the Shapelet of the current node is calculated, and the distance is compared with the optimal partition point to be divided into the left subtree or the right subtree. The above operation is repeated for each node until it is divided into a leaf node.

[0069] However, the original Shapelet method takes a long time, firstly because there are a large number of candidate subsequences, and secondly because it needs to calculate the information gain for all possible partition points.

[0070] In order to overcome the time-consuming problem of the original Shapelet algorithm in finding the optimal Shapelet at each node when building a decision tree, the Shapelet Transformation (ST) algorithm can be used. This algorithm is different from the traditional Shapelet method. It can efficiently identify all Shapelets by traversing the candidate subsequence set only once, so the computational efficiency of the ST algorithm is significantly improved. In addition, the ST algorithm can also be used in combination with different classification or regression algorithms to enhance model performance. The implementation of the ST algorithm is divided into two key steps:

[0071] (1) Finding Shapelets: This step is similar to the original Shapelet method, except that when it includes a subsequence in the candidate set, it first calculates the subsequence score based on a specific evaluation metric and puts both into the candidate set at the same time. After obtaining all candidate subsequences and scores, the ST algorithm removes candidate subsequences with high similarity and sorts them from high to low according to the scores. Finally, only the top k subsequences with the highest scores are retained as Shapelets, where k is a preset hyperparameter.

[0072] (2) Extracting time series features: After selecting k Shapelets, the ST algorithm will use them to extract the k-dimensional distance data of the time series. Suppose there is a time series , the ST algorithm can transform it into a new representation , thereby reducing the dimension of the initial time series. Once the distance data of the input time series is obtained, it can be combined with classification or regression algorithms to complete the subsequent work.

[0073] Based on the Shapelet transform, the Learning Time-series Shapelet (LTS) algorithm is introduced. This algorithm can directly learn the optimal Shapelets without searching all candidate objects. Its principle is divided into two steps: one is to make a rough initial guess for the Shapelets, and the other is to iteratively learn and optimize the Shapelets by minimizing the regression loss function. To this end, this paper defines a differentiable regression model for Shapelets, so that Shapelets can be updated by stochastic gradient descent by minimizing the regression loss. The flowchart is shown in the figure below. Figure 3 As shown:

[0074] The learning model is defined as follows:

[0075] (5)

[0076] In the formula, the minimum distance M is the new prediction operator after Shapelets transformation, is the linear weight parameter, is the bias term coefficient, To predict the approximate target value, the loss function is defined to minimize the difference between the predicted value and the actual value.

[0077] F is the regularized objective function, which minimizes the regression objective function by jointly learning the optimal Shapelet S and the optimal linear hyperplane W. The regularized objective function is defined as follows:

[0078] (6)

[0079] In order to calculate the derivative of the objective function, all relevant functions of the model must be differentiable. However, the minimum distance calculation function in formula (2) is not differentiable, and the partial derivative The time series T is not defined either. The differentiable approximation of the minimum function will be introduced below. i The distance between the jth segment and the kth Shapelet is defined as D i,k,j , as shown below:

[0080] (7)

[0081] The differentiable approximation of the minimum function is a commonly used softmin function, as shown in the following formula. α controls the accuracy of the function. When α→-∞, the soft minimum function value is close to the true minimum value.

[0082] (8)

[0083] In this method, stochastic gradient descent is used for optimization. The decomposed objective function Fi is given in the following formula. The objective decomposition of equation (6) is the single instance loss of each time series.

[0084] (9)

[0085] The learning algorithm updates Shapelets by calculating the gradient of the objective function relative to Shapelets. In the i-th time series, the gradient value of the l-th point of the k-th Shapelet is as follows:

[0086] (10)

[0087] In the formula, the predicted target gradient value of the minimum distance is defined as follows:

[0088] (11)

[0089] In addition, the gradient value of the overall minimum distance relative to the subsequence distance is as follows, and the gradient value of the subsequence distance relative to the Shapelet is as shown in formula (13).

[0090] (12)

[0091] (13)

[0092] The hyperplane weight W can also be learned by stochastic gradient descent to minimize the regression objective. Equation (14) represents the partial gradient for updating each weight Wk, and Equation (15) represents the bias term W0.

[0093] (14)

[0094] (15)

[0095] After calculating the optimal Shapelets, they can be converted into corresponding distance data to obtain a distance data set.

[0096] Step 103, performing regression learning on the distance data set and the frequency response lowest point data set to establish a lowest point prediction model;

[0097] In the embodiment of the present invention, support vector regression (SVR) may be used to perform regression learning on the distance data set and the frequency response lowest point data set to establish a lowest point prediction model.

[0098] The regression learning process of support vector regression is as follows:

[0099] When the linear regression function For sample data When fitting, the task of machine learning is to find a function f(x) that not only deviates from the target value y by less than the insensitive loss function , and the VC dimension of the function is minimal. The optimization goal is:

[0100] (16)

[0101] use After that, the minimization constraint changes to:

[0102] (17)

[0103] Taking into account the error caused by fitting, the slack variable is introduced , the above formula becomes:

[0104] (18)

[0105] Introduce the penalty coefficient C to indicate the excess error The penalty degree of the sample is changed to minimize the value of the following formula.

[0106] (19)

[0107] At this point, the regression problem becomes a constrained optimization problem:

[0108] (20)

[0109] (twenty one)

[0110] Convert the above equation into a quadratic programming problem and establish the Lagrange equation as follows:

[0111] (twenty two)

[0112] Find the partial derivative of the above formula:

[0113]

[0114] The dual optimization problem can be obtained:

[0115] (twenty three)

[0116] (twenty four)

[0117] In the formula, is the Lagrange multiplier, solving the above quadratic programming problem, we can get:

[0118] (25)

[0119] Under normal circumstances, Most of them will be equal to zero, and the samples corresponding to the small part that is not zero are the support vectors, so we have:

[0120] (26)

[0121] In the formula, SVs is the support vector set, and the value of b can be calculated using any sample:

[0122] (27)

[0123] Finally, f(x) can be expressed as:

[0124] (28)

[0125] Linear SVR is extended to nonlinear SVR through kernel functions. The basic idea is to map the input to a high-dimensional linearly separable feature space through nonlinear transformation, and use linear SVM in this space. Commonly used kernel functions are:

[0126] (1) Polynomial kernel function

[0127] (2) Radial basis kernel function

[0128] (3) Sigmoid kernel function

[0129] Among them, the radial basis kernel function is the most commonly used, and it is also selected as the kernel function in this chapter. Assume that the nonlinear model is:

[0130] (29)

[0131] Converting to a quadratic programming problem and establishing the Lagrange equation is as follows:

[0132] (30)

[0133] (31)

[0134] Introduce the kernel function, let , and finally get:

[0135] (32)

[0136] Support vector regression learning is similar to neural network in form. Each intermediate node corresponds to a support vector, and the output is a linear combination of the support vector and the bias. Its weight is the Lagrange multiplier corresponding to the support vector, such as Figure 4 shown.

[0137] Step 104, when a fault is detected in the power system network, obtaining time series data of measured frequency samples of the power system network;

[0138] Step 105, converting the measured frequency sample time series data into target distance data;

[0139] In an embodiment of the present invention, when a fault is detected in the power system network, the measured frequency sample time series data of the power system network can be obtained, and then the measured frequency sample time series data can be converted into target distance data.

[0140] In one example, the step of converting the measured frequency sample time series data into target distance data includes:

[0141] S51, obtaining initial time series characteristics of measured frequency sample time series data;

[0142] S52, iteratively optimize the initial time series features by minimizing the regression loss function to generate the optimal time series feature Shapelets of the measured frequency sample time series data;

[0143] S53, extracting multidimensional distance data of the time series data according to the optimal time series feature Shapelets as target distance data corresponding to the time series data.

[0144] In a specific implementation, the calculation process of the optimal time series feature Shapelets and the target distance data can refer to the LTS algorithm in steps 101-102, which will not be repeated here.

[0145] Step 106: Input the target distance data into the lowest point prediction model to obtain the frequency response lowest point prediction result of the measured frequency sample data.

[0146] In the specific implementation, once a fault occurs in the power system network, the measured frequency sample time series data is collected within a very short time after the fault, converted into distance data through Shapelet transformation and input into the trained regression model, so as to predict the lowest point of the frequency response in this scenario.

[0147] The present invention retains the time series characteristics of the time series data by converting the time series data into distance data, thereby improving the prediction accuracy of the lowest point of the frequency response of the power system.

[0148] For ease of understanding, the embodiments of the present invention are described below by using specific examples:

[0149] like Figure 5 As shown, Figure 5 A prediction flow chart of the lowest point of frequency response of a power system provided by an embodiment of the present invention. The prediction process is divided into three parts: case generation and data preprocessing, regression model training and frequency response lowest point prediction.

[0150] Case generation and data preprocessing: Model the power system network in RTDS, perform transient simulation to collect frequency data sets, and extract the lowest point of its frequency response. Use Shapelet transformation to convert the time series data set into a distance data set.

[0151] Regression model training: The distance data set and its corresponding lowest point data set are regressed through SVM to establish a lowest point prediction model.

[0152] When a fault is detected in the power system, the measured frequency sample time series data is collected within a very short time after the fault, converted into distance data through Shapelet transformation and input into the trained regression model, so as to predict the lowest point of the frequency response in this scenario.

[0153] See also Figure 6 , Figure 6 A structural block diagram of a device for predicting the lowest point of frequency response of a power system provided in an embodiment of the present invention.

[0154] An embodiment of the present invention provides a device for predicting the lowest point of frequency response of a power system, comprising:

[0155] The transient simulation module 601 is used to perform transient simulation on the power system network, collect frequency data sets, and extract the frequency response minimum point data sets in the frequency data sets; the frequency data sets include a plurality of time series data;

[0156] The distance data set acquisition module 602 is used to convert the time series data into distance data to obtain a distance data set;

[0157] The lowest point prediction model building module 603 is used to perform regression learning on the distance data set and the frequency response lowest point data set to build a lowest point prediction model;

[0158] The measured frequency sample time series data acquisition module 604 is used to acquire the measured frequency sample time series data of the power system network when a fault is detected in the power system network;

[0159] The target distance data conversion module 605 is used to convert the measured frequency sample time series data into the target distance data;

[0160] The frequency response lowest point prediction module 606 is used to input the target distance data into the lowest point prediction model to obtain the frequency response lowest point prediction result of the measured frequency sample data.

[0161] In the embodiment of the present invention, the target distance data conversion module 605 includes:

[0162] The optimal time series feature Shapelets extraction submodule is used to extract the optimal time series feature Shapelets from the measured frequency sample time series data according to the learning time series algorithm LTS;

[0163] The target distance data conversion submodule is used to extract the multi-dimensional distance data of the time series data according to the optimal time series feature Shapelets as the target distance data corresponding to the time series data.

[0164] In the embodiment of the present invention, the lowest point prediction model building module 603 includes:

[0165] The lowest point prediction model establishment submodule is used to use support vector regression SVR to perform regression learning on the distance data set and the frequency response lowest point data set to establish the lowest point prediction model.

[0166] In the embodiment of the present invention, the optimal time series feature Shapelets extraction submodule includes:

[0167] An initial time series feature acquisition unit, used to acquire initial time series features of measured frequency sample time series data;

[0168] The optimal time series feature Shapelets generation unit is used to iteratively optimize the initial time series features by minimizing the regression loss function to generate the optimal time series feature Shapelets of the measured frequency sample time series data.

[0169] An embodiment of the present invention further provides an electronic device, the device comprising a processor and a memory:

[0170] The memory is used to store the program code and transmit the program code to the processor;

[0171] The processor is used to execute the method for predicting the lowest point of frequency response of a power system according to the instructions in the program code.

[0172] The embodiment of the present invention further provides a computer-readable storage medium, which is used to store program codes, and the program codes are used to execute the method for predicting the lowest point of the frequency response of a power system according to the embodiment of the present invention.

[0173] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0174] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0175] It will be appreciated by those skilled in the art that the embodiments of the present invention may be provided as methods, devices, or computer program products. Therefore, the embodiments of the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the embodiments of the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0176] The embodiments of the present invention are described with reference to the flowcharts and / or block diagrams of the methods, terminal devices (systems), and computer program products according to the embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing terminal device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0177] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing terminal device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0178] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device so that a series of operating steps are executed on the computer or other programmable terminal device to produce a computer-implemented process, thereby providing instructions for executing on the computer or other programmable terminal device to implement the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0179] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the embodiments of the present invention.

[0180] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or terminal device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or terminal device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or terminal device including the elements.

[0181] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting the lowest point of frequency response of a power system, characterized in that: include: Performing transient simulation on the power system network, collecting frequency data sets, and extracting frequency response lowest point data sets from the frequency data sets; The frequency data set includes a plurality of time series data; Converting the time series data into distance data to obtain a distance data set; Performing regression learning on the distance data set and the frequency response lowest point data set to establish a lowest point prediction model; When a fault is detected in the power system network, obtaining time series data of measured frequency samples of the power system network; Convert measured frequency sample time series data into target distance data; The target distance data is input into the lowest point prediction model to obtain the frequency response lowest point prediction result of the measured frequency sample data.

2. The method according to claim 1, characterized in that The step of converting the measured frequency sample time series data into target distance data comprises: Extracting the optimal time series feature Shapelets from the measured frequency sample time series data according to the learning time series algorithm LTS; The multidimensional distance data of the time series data is extracted according to the optimal time series feature Shapelets as the target distance data corresponding to the time series data.

3. The method according to claim 1, characterized in that The step of performing regression learning on the distance data set and the frequency response lowest point data set to establish a lowest point prediction model comprises: Support vector regression (SVR) is used to perform regression learning on the distance data set and the frequency response lowest point data set to establish a lowest point prediction model.

4. The method according to claim 2, characterized in that: The step of extracting the optimal time series feature Shapelets in the measured frequency sample time series data according to the learning time series algorithm LTS includes: Obtaining initial time series characteristics of the measured frequency sample time series data; The initial time series features are iteratively optimized by minimizing the regression loss function to generate optimal time series feature Shapelets of the measured frequency sample time series data.

5. A device for predicting the lowest point of frequency response of a power system, characterized in that: include: A transient simulation module is used to perform transient simulation on the power system network, collect frequency data sets, and extract the frequency response minimum point data set in the frequency data sets; The frequency data set includes a plurality of time series data; A distance data set acquisition module, used for converting the time series data into distance data to obtain a distance data set; A lowest point prediction model building module is used to perform regression learning on the distance data set and the frequency response lowest point data set to build a lowest point prediction model; A measured frequency sample time series data acquisition module is used to acquire the measured frequency sample time series data of the power system network when a fault is detected in the power system network; A target distance data conversion module is used to convert the measured frequency sample time series data into target distance data; The frequency response lowest point prediction module is used to input the target distance data into the lowest point prediction model to obtain the frequency response lowest point prediction result of the measured frequency sample data.

6. The device according to claim 5, characterized in that The target distance data conversion module comprises: The optimal time series feature Shapelets extraction submodule is used to extract the optimal time series feature Shapelets in the measured frequency sample time series data according to the learning time series algorithm LTS; The target distance data conversion submodule is used to extract the multidimensional distance data of the time series data according to the optimal time series feature Shapelets as the target distance data corresponding to the time series data.

7. The device according to claim 5, characterized in that The lowest point prediction model establishment module includes: The lowest point prediction model establishment submodule is used to use support vector regression SVR to perform regression learning on the distance data set and the frequency response lowest point data set to establish a lowest point prediction model.

8. The device according to claim 6, characterized in that The optimal time series feature Shapelets extraction submodule includes: An initial time series feature acquisition unit, used to acquire the initial time series features of the measured frequency sample time series data; The optimal time series feature Shapelets generating unit is used to iteratively optimize the initial time series features by minimizing the regression loss function to generate the optimal time series feature Shapelets of the measured frequency sample time series data.

9. An electronic device, characterized in that: The device comprises a processor and a memory: The memory is used to store program code and transmit the program code to the processor; The processor is used to execute the method for predicting the lowest point of frequency response of a power system according to any one of claims 1 to 4 according to the instructions in the program code.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium is used to store program codes, and the program codes are used to execute the method for predicting the lowest point of frequency response of a power system as described in any one of claims 1-4.

Images