Power system disturbance lowest frequency prediction method based on physical information embedded layer

Abstract

The invention discloses a power system disturbance lowest frequency prediction method based on a physical information embedded layer, and the method comprises the steps: dividing a system frequency response physical model lowest frequency expression into different parts, constructing physical information embedded layer features which comprise first feature information, first weight information and first bias information, further obtaining a physical information embedded layer output function; and normalizing the output function of the physical information embedded layer and the system operation characteristic data, merging the two data, embedding the two data as an input layer into the long-short-term memory network, and constructing a novel power system lowest frequency prediction model based on the physical information embedded layer. According to the method, the frequency prediction physical model and the deep learning algorithm are deeply fused, the physical knowledge mining capability of the data-driven model is enhanced, and the lowest frequency of the system after disturbance can be rapidly and accurately predicted.

Description

Technical Field

[0001] The present invention relates to the field of new energy power systems, and in particular to a method for predicting the minimum frequency of power system disturbances based on a physical information embedded layer. Background Art

[0002] With the large-scale grid connection of renewable energy, a large number of synchronous generators with rotational inertia are replaced by renewable energy sources, the inertia of the power system decreases, and the primary frequency regulation capacity weakens. In this context, the risk of frequency instability after power system disturbances will increase.

[0003] At present, the post-disturbance frequency analysis methods of power systems are mainly divided into physical model methods, data-driven methods and physical-data fusion methods. The new power system has a variety of frequency regulation units and a relatively complex structure. Under this background, the physical model method is difficult to balance the calculation accuracy and efficiency. In recent years, the rise of artificial intelligence has provided a feasible data-driven method for predicting the frequency response of power systems. However, since the data-driven method only relies on a large amount of training data for modeling, the lack of physical knowledge makes it impossible to further improve the prediction accuracy of the data-driven model. The physical-data-driven fusion method combines the prediction results of the physical model and the data-driven model to achieve complementary advantages. Some deep fusion methods can also fully explore the physical knowledge hidden in the physical model, and the mined physical knowledge can further improve the prediction accuracy of the data-driven model. Using the physical-data fusion method for power system frequency prediction can achieve more accurate power system frequency prediction and better meet the frequency prediction needs. Summary of the invention

[0004] In view of this, the purpose of the present invention is to propose a method for predicting the minimum frequency of power system disturbances based on a physical information embedded layer. By constructing a physical information embedded layer, the frequency prediction physical model and the deep learning algorithm are deeply integrated, thereby enhancing the ability of the data-driven model to mine physical knowledge, and being able to quickly and accurately predict the minimum frequency of the system after the disturbance.

[0005] In order to achieve the above technical objectives, the technical solution adopted by the present invention is:

[0006] A method for predicting the minimum frequency of disturbances in a power system based on a physical information embedded layer, the method comprising:

[0007] Acquire multiple disturbance information, each disturbance information is obtained by simulating a certain fault condition in the power system through a simulation model, the disturbance information includes system operation information and actual value of frequency response, and the system operation information includes system operation data before and after the disturbance caused by the fault condition;

[0008] A system frequency response physical model is established according to the plurality of disturbance information, and a minimum frequency function is obtained according to the system frequency response physical model;

[0009] Converting the lowest frequency function into a physical information embedded layer feature, wherein the physical information embedded layer feature includes first feature information, first weight information, and first bias information;

[0010] Calculate the output function of the physical information embedded layer according to the physical information embedded layer characteristics;

[0011] The output function of the physical information embedding layer and the system operation information are merged and input into the long short-term memory network model to obtain the minimum frequency prediction model based on the long short-term memory network model;

[0012] and, constructing a frequency response database based on the disturbance information;

[0013] The frequency response database is used as sample data of the minimum frequency prediction model and is trained to obtain a trained minimum frequency prediction model.

[0014] In some embodiments, a system frequency response physical model is established according to a plurality of disturbance information, and obtaining a minimum frequency function according to the system frequency response physical model further comprises:

[0015] Identifying parameter information of a system frequency response physical model according to a plurality of disturbance information, the parameter information including a system frequency response function coefficient, an angular frequency, an oscillation decay time constant, a damping decay time constant, and an initial phase angle;

[0016] A system frequency response physical model is constructed according to the parameter information. The lowest frequency is substituted into the system frequency response physical model to obtain a lowest frequency function. The lowest frequency function is expressed by formula (1). Formula (1) is as follows:

[0017]

[0018] In formula (1), f minSFR is the lowest frequency calculated by the physical model of the system frequency response, t minSFR is the extreme value time calculated by the physical model of the system frequency response, f N is the rated frequency, P d is the active disturbance, T1 is the damping decay time constant, T2 is the oscillation decay time constant, ω2 is the angular frequency, θ is the initial phase angle, C0, C1, C2 are the system frequency response function coefficients.

[0019] In some embodiments, identifying parameter information of a system frequency response physical model according to a plurality of disturbance information includes:

[0020] Given a sampling time, the frequency value corresponding to the sampling time is calculated according to the system frequency response time domain function, which is expressed by formula (2). Formula (2) is as follows:

[0021]

[0022] In formula (2), t is the given sampling time, f t is the frequency value at time t;

[0023] The least square method is used to fit the actual value of the frequency response in formula (2). The least square method uses the sum of squared errors as the objective function, which is expressed by formula (3). Formula (3) is as follows:

[0024]

[0025] In formula (3), is parameter information, represents the first system frequency dynamic response data, which is calculated by the system frequency response physical model. r，k It represents the second system frequency dynamic response data, which is obtained by simulation of the simulation model, and N1 represents the number of sampling points.

[0026] In some embodiments, converting the lowest frequency function into a physical information embedded layer feature, where the physical information embedded layer feature includes first feature information, first weight information, and first bias information, includes:

[0027] In formula (1), and P d The first feature information converted into the physical information embedded layer feature is expressed by formula (4), which is as follows:

[0028]

[0029] In formula (4), X PI It is the first characteristic information of the embedded layer characteristics of the physical information;

[0030] C0, C1, and C2 in formula (1) are converted into the first weight information of the physical information embedded layer feature, which is expressed by formula (5):

[0031] W PI =[w0,w1,w2]=[C0f N , C1f N , C2f N ],

[0032] In formula (5), W PI is the first weight information of the embedded layer feature of the physical information, w0, w1, w2 are three components in the first weight information of the embedded layer feature of the physical information;

[0033] And, replace f in formula (1) N The first bias information converted into the physical information embedded layer feature is expressed by formula (6):

[0034] b PI =[f N ],

[0035] In formula (6), b PI It is the first bias information of the embedded layer feature of the physical information.

[0036] In some embodiments, the output function of the physical information embedding layer is calculated according to the physical information embedding layer characteristics, which is expressed by formula (7). Formula (7) is as follows:

[0037] f PIL =W PI X PI +b PI ,

[0038] In formula (7), f PIL is the output function of the physical information embedding layer.

[0039] In some embodiments, the output function of the physical information embedding layer and the system operation information are merged and input into the long short-term memory network model, and the minimum frequency prediction model based on the long short-term memory network model is obtained, including:

[0040] Normalizing the output function of the physical information embedded layer and the system operation information to obtain first physical information embedded layer data and first system operation information;

[0041] Merging the first physical information embedded layer data and the first system operation information to obtain a first data set;

[0042] The first data set is used as the input layer of the long short-term memory network model to obtain the lowest frequency prediction model.

[0043] In some embodiments, the frequency response database is used as sample data of the minimum frequency prediction model and trained to obtain a trained minimum frequency prediction model, including:

[0044] Divide the disturbance information in the frequency response database into a training set and a test set according to a certain ratio;

[0045] Normalizing the training set and the test set to obtain a second data set, the second data set includes a processed test set and a processed training set, the processed training set includes the first lowest frequency actual value, and the processed test set includes the second lowest frequency actual value;

[0046] The first lowest frequency actual value is used as a training label, and the Adam algorithm is used to update the second weight information and the second bias information of the long short-term memory network model;

[0047] Performing supervised training on the long short-term memory network model using the processed training set to obtain a first prediction result, where the first prediction result is a first lowest frequency prediction value of the processed training set;

[0048] Adjusting the hyperparameters of the LSTM network model according to the first prediction result;

[0049] and, using the processed test set to verify the trained long short-term memory network model to obtain a second prediction result, where the second prediction result is a second lowest frequency prediction value of the processed test set;

[0050] Performing a denormalization process on the second prediction result to obtain a third lowest frequency prediction value;

[0051] Calculate the lowest frequency error of the long short-term memory network model according to the second lowest frequency actual value and the third lowest frequency predicted value;

[0052] Determining whether the minimum frequency error is within a preset threshold value;

[0053] If yes, it means that the long short-term memory network model training is complete;

[0054] If not, the hyperparameters of the long short-term memory network model are updated, and the third lowest frequency prediction value is updated synchronously, and the lowest frequency error is recalculated until the lowest frequency error is within the range of the preset threshold.

[0055] In some embodiments, the system operation data includes each bus voltage and phase angle in the power system, each generator frequency, mechanical power and electromagnetic power, load active power and system power deficit.

[0056] In some embodiments, the simulation model is PSD-BPA.

[0057] In some embodiments, the power system is an extended IEEE-39 node system.

[0058] By adopting the above technical solution, the present invention has the following beneficial effects compared with the prior art:

[0059] The present invention proposes a method for predicting the minimum frequency of disturbance in a power system based on a physical information embedded layer, which makes full use of physical information, converts the terms containing active disturbance and extreme time in the minimum frequency function of the system frequency response physical model into input features of the neural network, converts the terms containing the parameters to be identified in the expression into weights of the neural network, and constructs a physical information embedded layer. By establishing a physical model of system frequency response, the minimum frequency function is converted into physical information embedded layer features, and it is combined with system operation information to input the prediction model. The proposed physical information embedded layer features can correct the physical model parameters through network training, mine more accurate physical knowledge, and obtain better preliminary prediction results than the original physical model. Compared with the pure data-driven method, the corrected physical model can more accurately capture the dynamic change law of the system frequency, and the preliminary prediction results significantly improve the minimum frequency prediction performance of the fusion model. By using time domain simulation software to batch simulate various fault conditions of the power system, a rich frequency response database covering various fault conditions is constructed, which provides strong support for subsequent model training and optimization and enhances the generalization ability of the model. In addition, this embodiment fuses the output of the physical information embedded layer with the system operation information as the input of the long short-term memory network, which not only retains the physical mechanism information, but also utilizes the long short-term memory network's modeling ability for time series data, which can better characterize the complex characteristics of the dynamic response of the power system frequency, and the prediction performance is more superior. On the other hand, based on the verification of the extended IEEE-39 node system, it can be applied to large-scale and complex new power systems, and provides reliable prediction support for power system frequency stability analysis and control. In short, the method proposed in this embodiment integrates physical mechanism and data-driven modeling technology, and has significant advantages in improving prediction accuracy and enhancing generalization ability, and has important application value for dynamic analysis and control of power systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] 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 work.

[0061] Figure 1 It is a schematic diagram of steps S101 to S106 of the method for predicting the lowest frequency of disturbance in a power system based on a physical information embedded layer according to a specific implementation manner;

[0062] Figure 2 It is a schematic diagram of steps S201 to S210 of the method for predicting the lowest frequency of disturbance in a power system based on a physical information embedded layer according to a specific implementation manner;

[0063] Figure 3 is a schematic diagram of the structure of the minimum frequency prediction model based on the physical information embedded layer described in the specific implementation method;

[0064] Figure 4 It is a flowchart of a method for predicting the minimum frequency of power system disturbances based on a physical information embedded layer as described in a specific implementation manner. DETAILED DESCRIPTION

[0065] The present invention will be further described in detail below in conjunction with the accompanying drawings and examples. It is particularly noted that the following examples are only used to illustrate the present invention, but are not intended to limit the scope of the present invention. Similarly, the following examples are only partial embodiments of the present invention rather than all embodiments, and all other embodiments obtained by those of ordinary skill in the art without creative work are within the scope of protection of the present invention.

[0066] The present invention provides a method for predicting the minimum frequency of power system disturbance based on a physical information embedded layer. By constructing a physical information embedded layer, the frequency prediction physical model and the deep learning algorithm are deeply integrated, thereby enhancing the ability of the data-driven model to mine physical knowledge, and being able to quickly and accurately predict the minimum frequency of the system after the disturbance.

[0067] See also Figure 1 and Figure 4 This embodiment provides a method for predicting the minimum frequency of power system disturbance based on a physical information embedded layer, the method comprising:

[0068] S101, obtaining a plurality of disturbance information, each disturbance information being obtained by simulating a certain fault condition in the power system through a simulation model, the disturbance information including system operation information and an actual value of a frequency response, the system operation information including system operation data before and after the disturbance caused by the fault condition;

[0069] S102, establishing a system frequency response physical model according to the plurality of disturbance information, and obtaining a minimum frequency function according to the system frequency response physical model;

[0070] S103, converting the lowest frequency function into a physical information embedded layer feature, where the physical information embedded layer feature includes first feature information, first weight information, and first bias information;

[0071] S104, calculating the output function of the physical information embedded layer according to the physical information embedded layer characteristics;

[0072] S105, merging the output function of the physical information embedded layer and the system operation information, and inputting them into the long short-term memory network model to obtain a minimum frequency prediction model based on the long short-term memory network model;

[0073] and, constructing a frequency response database based on the disturbance information;

[0074] S106: Use the frequency response database as sample data of the minimum frequency prediction model and perform training to obtain a trained minimum frequency prediction model.

[0075] In this embodiment, the method for predicting the lowest frequency of power system disturbance based on the physical information embedded layer has the following specific steps:

[0076] Step 1: Use time domain simulation software to batch simulate various fault conditions in the new power system to generate disturbance information of the power system. Select system operation data before and after the disturbance and actual frequency response values ​​to build a frequency response database;

[0077] Step 2: Use the actual frequency response value obtained in step 1 and the system operation data to identify the system frequency response physical model parameters, establish the system frequency response physical model, and obtain the lowest frequency function after the system disturbance. Divide the lowest frequency function into different parts, construct the physical information embedded layer features, including the first feature information, the first weight information and the first bias information, and further calculate the physical information embedded layer output function;

[0078] Step 3: Reconstruct the output function of the physical information embedded layer and the system operation information obtained in step 1, normalize them and merge them, embed them into the long short-term memory network as the input layer, and build a power system minimum frequency prediction model based on the physical information embedded layer. With the support of the frequency response database, the proposed power system minimum frequency prediction model based on the physical information embedded layer is trained;

[0079] Step 4: Input the power system operation data obtained online into the frequency prediction model trained in step 3 to realize the minimum frequency prediction of the new power system.

[0080] This embodiment proposes a method for predicting the minimum frequency of power system disturbance based on a physical information embedded layer, which makes full use of physical information, converts the terms containing active disturbance and extreme time in the minimum frequency function of the system frequency response physical model into input features of the neural network, converts the terms containing the parameters to be identified in the expression into the weights of the neural network, and constructs a physical information embedded layer. By establishing a physical model of system frequency response, the minimum frequency function is converted into a physical information embedded layer feature, and it is combined with the system operation information to input the prediction model. The proposed physical information embedded layer feature can correct the physical model parameters through network training, mine more accurate physical knowledge, and obtain better preliminary prediction results than the original physical model. Compared with the pure data-driven method, the corrected physical model can more accurately capture the dynamic change law of the system frequency, and the preliminary prediction results significantly improve the minimum frequency prediction performance of the fusion model. By using time domain simulation software to batch simulate various fault conditions of the power system, a rich frequency response database covering various fault conditions is constructed, which provides strong support for subsequent model training and optimization and enhances the generalization ability of the model. In addition, this embodiment fuses the output of the physical information embedded layer with the system operation information as the input of the long short-term memory network, which not only retains the physical mechanism information, but also utilizes the long short-term memory network's modeling ability for time series data, which can better characterize the complex characteristics of the dynamic response of the power system frequency, and the prediction performance is more superior. On the other hand, based on the verification of the extended IEEE-39 node system, it can be applied to large-scale and complex new power systems, and provides reliable prediction support for power system frequency stability analysis and control. In short, the method proposed in this embodiment integrates physical mechanism and data-driven modeling technology, and has significant advantages in improving prediction accuracy and enhancing generalization ability, and has important application value for dynamic analysis and control of power systems.

[0081] In some embodiments, a system frequency response physical model is established according to a plurality of disturbance information, and obtaining a minimum frequency function according to the system frequency response physical model further comprises:

[0082] Identifying parameter information of a system frequency response physical model according to a plurality of disturbance information, the parameter information including a system frequency response function coefficient, an angular frequency, an oscillation decay time constant, a damping decay time constant, and an initial phase angle;

[0083] A system frequency response physical model is constructed according to the parameter information. The lowest frequency is substituted into the system frequency response physical model to obtain a lowest frequency function. The lowest frequency function is expressed by formula (1). Formula (1) is as follows:

[0084]

[0085] In formula (1), f minSFRis the lowest frequency calculated by the physical model of the system frequency response, t minSFR is the extreme value time calculated by the physical model of the system frequency response, f N is the rated frequency, P d is the active disturbance, T1 is the damping decay time constant, T2 is the oscillation decay time constant, ω2 is the angular frequency, θ is the initial phase angle, C0, C1, C2 are the system frequency response function coefficients.

[0086] In this embodiment, the actual value of the frequency response obtained by the above method and the system operation data are used to identify the system frequency response physical model parameters, establish the system frequency response physical model, and obtain the lowest frequency expression after the system disturbance. The lowest frequency expression is divided into different parts, and the physical information embedded layer features are constructed, including the first feature information, the first weight information and the first bias information, and the physical information embedded layer output function is further calculated.

[0087] The lowest frequency time domain expression of the system frequency response physical model is shown in formula (1):

[0088]

[0089] In formula (1), f minSFR is the lowest frequency calculated by the physical model of the system frequency response, t minSFR is the extreme value time calculated by the physical model of the system frequency response, f N is the rated frequency, P d is the active disturbance, T1 is the damping decay time constant, T2 is the oscillation decay time constant, ω2 is the angular frequency, θ is the initial phase angle, C0, C1, C2 are the system frequency response function coefficients.

[0090] In some embodiments, identifying parameter information of a system frequency response physical model according to a plurality of disturbance information includes:

[0091] Given a sampling time, the frequency value corresponding to the sampling time is calculated according to the system frequency response time domain function, which is expressed by formula (2). Formula (2) is as follows:

[0092]

[0093] In formula (2), t is the given sampling time, f t is the frequency value at time t;

[0094] The least square method is used to fit the actual value of the frequency response in formula (2). The least square method uses the sum of squared errors as the objective function, which is expressed by formula (3). Formula (3) is as follows:

[0095]

[0096] In formula (3), is parameter information, represents the first system frequency dynamic response data, which is calculated by the system frequency response physical model. r，k It represents the second system frequency dynamic response data, which is obtained by simulation of the simulation model, and N1 represents the number of sampling points.

[0097] In this embodiment, the values ​​of C0, C1, C2, T1, T2, ω2, and θ are obtained by least squares identification. The specific identification process is as follows:

[0098] Given a sampling time t, the frequency value f corresponding to the sampling time t is calculated using the system frequency response time domain expression shown in formula (2): t .

[0099]

[0100] Use the least squares method to fit the actual frequency response data of the system and obtain the system frequency response model parameters The objective function of the least squares method uses the sum of squared errors, which is expressed as follows:

[0101]

[0102] In formula (3), represents the system frequency dynamic response data obtained by calculating the physical model (ie, the first system frequency dynamic response data); f r，k represents the system frequency dynamic response data obtained by full time domain simulation (ie, the second system frequency dynamic response data); N1 represents the number of sampling points.

[0103] In some embodiments, converting the lowest frequency function into a physical information embedded layer feature, where the physical information embedded layer feature includes first feature information, first weight information, and first bias information, includes:

[0104] In formula (1), and P d The first feature information converted into the physical information embedded layer feature is expressed by formula (4), which is as follows:

[0105]

[0106] In formula (4), X PI It is the first characteristic information of the embedded layer characteristics of the physical information;

[0107] C0, C1, and C2 in formula (1) are converted into the first weight information of the physical information embedded layer feature, which is expressed by formula (5):

[0108] W PI =[w0,w1,w2]=[C0f N , C1f N , C2f N ],

[0109] In formula (5), W PI is the first weight information of the embedded layer feature of the physical information, w0, w1, w2 are three components in the first weight information of the embedded layer feature of the physical information;

[0110] And, replace f in formula (1) N The first bias information converted into the physical information embedded layer feature is expressed by formula (6):

[0111] b PI =[f N ],

[0112] In formula (6), b PI It is the first bias information of the embedded layer feature of the physical information.

[0113] In this embodiment, formula (1) is divided into different parts to construct a physical information embedding layer. The specific steps are as follows:

[0114] First, extract the and P d The three parts contain active disturbance and extreme time information. For different samples, the values ​​of these three parts vary greatly and are not suitable for conversion into parameters of the neural network layer. Therefore, they are converted into input features X of the physical information embedding layer. PI (i.e., the first feature information of the physical information embedded layer feature), as shown in formula (4).

[0115]

[0116] The remaining part in formula (1) consists of some physical parameters C0, C1, C2 whose exact values ​​are unknown. These parts are converted into the first weight information W of the physical information embedding layer PI and the first bias information b PI , in order to use neural network training to correct unknown physical parameters and further explore the physical knowledge implicit in the parameters. The specific expressions of the constructed physical information embedded layer features and parameters are shown in (5)-(6).

[0117] W PI =[w0,w1,w2]=[C0f N , C1fN , C2f N ],

[0118] b PI =[f N ],

[0119] In formula (5), w0, w1, and w2 are three components of the first weight information of the embedded layer feature of the physical information.

[0120] In some embodiments, the output function of the physical information embedding layer is calculated according to the physical information embedding layer characteristics, which is expressed by formula (7). Formula (7) is as follows:

[0121] f PIL =W PI X PI +b PI ,

[0122] In formula (7), f PIL is the output function of the physical information embedding layer.

[0123] In some embodiments, the output function of the physical information embedding layer and the system operation information are merged and input into the long short-term memory network model, and the minimum frequency prediction model based on the long short-term memory network model is obtained, including:

[0124] Normalizing the output function of the physical information embedded layer and the system operation information to obtain first physical information embedded layer data and first system operation information;

[0125] Merging the first physical information embedded layer data and the first system operation information to obtain a first data set;

[0126] The first data set is used as the input layer of the long short-term memory network model to obtain the lowest frequency prediction model.

[0127] In this embodiment, the output of the physical information embedded layer and the system operation characteristic data obtained by the above method are reconstructed, the two are normalized and merged, and embedded into the long short-term memory network as the input layer to construct a new power system minimum frequency prediction model based on the physical information embedded layer. With the support of the frequency response database, the proposed new power system minimum frequency prediction model based on the physical information embedded layer is trained.

[0128] See also Figure 2 In some embodiments, the frequency response database is used as sample data of the minimum frequency prediction model and trained to obtain the trained minimum frequency prediction model, including:

[0129] S201, dividing the disturbance information in the frequency response database into a training set and a test set according to a certain ratio;

[0130] S202, normalizing the training set and the test set to obtain a second data set, where the second data set includes a processed test set and a processed training set, where the processed training set includes the first lowest frequency actual value, and the processed test set includes the second lowest frequency actual value;

[0131] S203, using the first lowest frequency actual value as a training label, and using the Adam algorithm to update the second weight information and the second bias information of the long short-term memory network model;

[0132] S204, using the processed training set to perform supervised training on the long short-term memory network model to obtain a first prediction result, where the first prediction result is a first lowest frequency prediction value of the processed training set;

[0133] S205, adjusting the hyperparameters of the long short-term memory network model according to the first prediction result;

[0134] and, using the processed test set to verify the trained long short-term memory network model to obtain a second prediction result, where the second prediction result is a second lowest frequency prediction value of the processed test set;

[0135] S206, performing a denormalization process on the second prediction result to obtain a third lowest frequency prediction value;

[0136] S207, calculating the lowest frequency error of the long short-term memory network model according to the second lowest frequency actual value and the third lowest frequency predicted value;

[0137] S208, determining whether the minimum frequency error is within a preset threshold range;

[0138] S209: If yes, it means that the long short-term memory network model training is completed;

[0139] S210: If not, update the hyperparameters of the long short-term memory network model, and simultaneously update the third lowest frequency prediction value, and recalculate the lowest frequency error until the lowest frequency error is within the range of a preset threshold.

[0140] In this embodiment, the specific training process can be combined with Figure 3 To understand:

[0141] The frequency response data set in the above method is divided into a training set and a test set according to a certain ratio, and the data is reconstructed and normalized. The processed physical information embedded layer output and system operation characteristic data are used as the input features of the long short-term memory network, and the processed lowest frequency actual value (i.e., the first lowest frequency actual value) is used as the training label. The Adam algorithm is used to update the second weight information and the second bias information of the long short-term memory network, and the model is supervised. At the same time, the hyperparameters of the model need to be adjusted according to the prediction results of the model on the training set (i.e., the first prediction results) to ensure that the trained model has good prediction performance.

[0142] Input the input features in the test set into the trained model to obtain the lowest frequency prediction value (i.e., the second lowest frequency prediction value), and denormalize the prediction result (i.e., the second prediction result). The lowest frequency prediction error of the model is calculated using the denormalized lowest frequency prediction value (i.e., the third lowest frequency prediction value) and the lowest frequency actual value (i.e., the second lowest frequency actual value) to evaluate the prediction accuracy of the model. If the prediction accuracy of the model meets the preset requirements, save the model. If not, reselect the hyperparameters for training.

[0143] In some embodiments, the system operation data includes each bus voltage and phase angle in the power system, each generator frequency, mechanical power and electromagnetic power, load active power and system power deficit.

[0144] In some embodiments, the simulation model is PSD-BPA.

[0145] In some embodiments, the power system is an extended IEEE-39 node system.

[0146] Furthermore, the following examples can be developed in combination with the above solutions:

[0147] The simulation software uses PSD-BPA, and the test system uses an extended IEEE-39 node system. The system has an initial total load of 6160.1MW, a rated frequency of 50Hz, a base capacity of 100MVA, and a base voltage of 345kV. Wind farms with rated powers of 571.2MW and 550.8MW are connected at nodes 36 and 37 of the system, respectively. A photovoltaic power station with a rated power of 70.35MW is incorporated at node 35 of the system, and the output power of the steam turbine generator at node 35 is reduced to 585MW.

[0148] PSD-BPA software loop calling program was written in Python to batch generate samples for model training. To simulate different operating scenarios of the system, the initial level of system load was set to vary between 80% and 105%, with a variation of 5%. Power disturbance was set by adding load on a single bus and changing the output of the generator. The disturbance range of the single bus load was 1.5%-6.5% of the initial load of the system, and the amplitude of each disturbance change was 0.5%. After removing the two loaded generators on buses 31 and 39, the wind farms connected to buses 36 and 37, and the photovoltaic stations and generators connected to bus 35, one of the remaining five generators was randomly selected to change the output each time. The output of the generator varied from 50% to 100% of the original output of the generator, and the amplitude of each output change was 10%. On this basis, 1434 samples were generated, and the sample set was divided into a training set and a test set at a ratio of 8:2. Grid search was used to select the long short-term memory (LSTM) model hyperparameters. The selected LSTM model had 3 layers, and the number of neurons in each layer was 100, 100, and 50, respectively.

[0149] In order to verify the superiority of the proposed minimum frequency prediction model based on the physical information embedded layer (named as PILSTM model), four fusion models were designed as comparison models with reference to different literatures. The LSTM_series fusion model, LSTM_parallel fusion model, and LSTM_guided fusion model were obtained by connecting the minimum frequency calculation results of the physical model and the LSTM model in different ways. In addition to the above three shallow fusion models, this paper also designed the LSTM_variable weight loss function model. On the basis of the traditional MSE loss function, this model adds weights that can reflect the error between the LSTM model prediction value and the physical model calculation value. In order to evaluate the prediction accuracy of the proposed model, the mean absolute error (MAE), mean absolute percentage error (MAPE) and root mean square error (RMSE) are introduced as evaluation indicators. At the same time, in order to facilitate the comparison of the prediction accuracy between the pure data-driven model and the fusion model, the prediction error of the LSTM model is also displayed. The prediction errors of different models are the average values ​​of the results of 5 random experiments, as shown in Table 1.

[0150] Table 1 Comparison of lowest frequency prediction errors of different models

[0151]

[0152] As shown in Table 1, the error of the physical-data fusion method is lower than that of the pure data-driven model LSTM. This shows that the addition of physical knowledge improves the prediction accuracy of the data-driven model. At the same time, there are large differences in the prediction accuracy of different fusion methods in Table 1. Among all the fusion models, the PILSTM model has the best prediction effect, which shows that the PILSTM model indirectly optimizes the parameters to be identified in the physical model through network training, introduces more accurate physical knowledge into the LSTM model, and obtains better minimum frequency prediction results.

[0153] By adopting the above technical solution, the present invention has the following beneficial effects compared with the prior art:

[0154] The present invention proposes a method for predicting the minimum frequency of disturbance in a power system based on a physical information embedded layer, which makes full use of physical information, converts the terms containing active disturbance and extreme time in the minimum frequency function of the system frequency response physical model into input features of the neural network, converts the terms containing the parameters to be identified in the expression into weights of the neural network, and constructs a physical information embedded layer. By establishing a physical model of system frequency response, the minimum frequency function is converted into physical information embedded layer features, and it is combined with system operation information to input the prediction model. The proposed physical information embedded layer features can correct the physical model parameters through network training, mine more accurate physical knowledge, and obtain better preliminary prediction results than the original physical model. Compared with the pure data-driven method, the corrected physical model can more accurately capture the dynamic change law of the system frequency, and the preliminary prediction results significantly improve the minimum frequency prediction performance of the fusion model. By using time domain simulation software to batch simulate various fault conditions of the power system, a rich frequency response database covering various fault conditions is constructed, which provides strong support for subsequent model training and optimization and enhances the generalization ability of the model. In addition, this embodiment fuses the output of the physical information embedded layer with the system operation information as the input of the long short-term memory network, which not only retains the physical mechanism information, but also utilizes the long short-term memory network's modeling ability for time series data, which can better characterize the complex characteristics of the dynamic response of the power system frequency, and the prediction performance is more superior. On the other hand, based on the verification of the extended IEEE-39 node system, it can be applied to large-scale and complex new power systems, and provides reliable prediction support for power system frequency stability analysis and control. In short, the method proposed in this embodiment integrates physical mechanism and data-driven modeling technology, and has significant advantages in improving prediction accuracy and enhancing generalization ability, and has important application value for dynamic analysis and control of power systems.

[0155] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0156] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, server, or network device, etc.) or a processor (processor) to perform all or part of the steps of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program code.

[0157] The above descriptions are only some embodiments of the present invention, and are not intended to limit the protection scope of the present invention. Any equivalent device or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for predicting the lowest frequency of disturbances in a power system based on a physical information embedded layer, characterized in that: The method comprises: Acquire multiple disturbance information, each of which is obtained by simulating a certain fault condition in the power system through a simulation model, the disturbance information includes system operation information and an actual value of a frequency response, and the system operation information includes system operation data before and after the disturbance caused by the fault condition; Establishing a system frequency response physical model according to the plurality of disturbance information, and obtaining a minimum frequency function according to the system frequency response physical model; Converting the lowest frequency function into a physical information embedded layer feature, wherein the physical information embedded layer feature includes first feature information, first weight information, and first bias information; Calculating an output function of the physical information embedded layer according to the physical information embedded layer characteristics; The output function of the physical information embedding layer and the system operation information are combined and input into a long short-term memory network model to obtain a minimum frequency prediction model based on the long short-term memory network model; and, constructing a frequency response database according to the disturbance information; The frequency response database is used as sample data of the minimum frequency prediction model and is trained to obtain a trained minimum frequency prediction model.

2. The method for predicting the lowest frequency of disturbances in a power system based on a physical information embedded layer according to claim 1, characterized in that: Establishing a system frequency response physical model according to the plurality of disturbance information, and obtaining a minimum frequency function according to the system frequency response physical model further comprises: Identifying parameter information of a system frequency response physical model according to the plurality of disturbance information, wherein the parameter information includes a system frequency response function coefficient, an angular frequency, an oscillation decay time constant, a damping decay time constant, and an initial phase angle; A system frequency response physical model is constructed according to the parameter information, and the lowest frequency is substituted into the system frequency response physical model to obtain a lowest frequency function, which is represented by formula (1). Formula (1) is as follows: In formula (1), f minSFR is the lowest frequency calculated by the physical model of the system frequency response, t minSFR is the extreme value time calculated by the physical model of the system frequency response, f N is the rated frequency, P d is the active disturbance, T1 is the damping decay time constant, T2 is the oscillation decay time constant, ω2 is the angular frequency, θ is the initial phase angle, C0, C1, C2 are the system frequency response function coefficients.

3. The method for predicting the lowest frequency of disturbances in a power system based on a physical information embedded layer according to claim 2, characterized in that: The parameter information of the physical model of the frequency response of the system is identified according to the plurality of disturbance information, including: Given a sampling time, the frequency value corresponding to the sampling time is calculated according to the system frequency response time domain function, which is expressed by formula (2). The formula (2) is as follows: In formula (2), t is the given sampling time, f t is the frequency value at time t; The actual value of the frequency response in the formula (2) is fitted using the least square method, which uses the sum of squared errors as the objective function and is expressed by formula (3). Formula (3) is as follows: In formula (3), is parameter information, represents the first system frequency dynamic response data, which is calculated by the system frequency response physical model, r，k represents the second system frequency dynamic response data, which is obtained by simulation of the simulation model, and N1 represents the number of sampling points.

4. The method for predicting the lowest frequency of disturbances in a power system based on a physical information embedded layer according to claim 2, characterized in that: Converting the lowest frequency function into a physical information embedded layer feature, wherein the physical information embedded layer feature includes first feature information, first weight information, and first bias information, including: In formula (1), and P d The first feature information converted into the physical information embedded layer feature is expressed by formula (4), which is as follows: In formula (4), X PI The first characteristic information of the embedded layer characteristic of the physical information; C0, C1, and C2 in formula (1) are converted into the first weight information of the physical information embedded layer feature, which is expressed by formula (5): IN PI =[w0,w1,w2]=[C0f N ,C1f N ,C2f N ], In formula (5), W PI is the first weight information of the embedded layer feature of the physical information, w0, w1, w2 are three components in the first weight information of the embedded layer feature of the physical information; And, replace f in formula (1) N The first bias information converted into the physical information embedded layer feature is expressed by formula (6): b PI =[f N ], In formula (6), b PI It is the first bias information of the embedded layer feature of the physical information.

5. The method for predicting the lowest frequency of disturbances in a power system based on a physical information embedded layer according to claim 4, characterized in that: The output function of the physical information embedding layer is calculated according to the physical information embedding layer characteristics, which is expressed by formula (7). Formula (7) is as follows: f PIL =W PI X PI +b PI , In formula (7), f PIL is the output function of the physical information embedding layer.

6. The method for predicting the lowest frequency of disturbances in a power system based on a physical information embedded layer according to claim 5, characterized in that: The output function of the physical information embedding layer and the system operation information are combined and input into the long short-term memory network model to obtain the lowest frequency prediction model based on the long short-term memory network model, including: Normalizing the output function of the physical information embedded layer and the system operation information to obtain first physical information embedded layer data and first system operation information; Merging the first physical information embedded layer data and the first system operation information to obtain a first data set; The first data set is used as the input layer of the long short-term memory network model to obtain a minimum frequency prediction model.

7. The method for predicting the lowest frequency of disturbances in a power system based on a physical information embedded layer according to claim 6, characterized in that: The frequency response database is used as sample data of the minimum frequency prediction model and trained to obtain a trained minimum frequency prediction model, which includes: Divide the disturbance information in the frequency response database into a training set and a test set according to a certain ratio; Normalizing the training set and the test set to obtain a second data set, wherein the second data set includes a processed test set and a processed training set, wherein the processed training set includes the first lowest frequency actual value, and the processed test set includes the second lowest frequency actual value; Using the first lowest frequency actual value as a training label, and using the Adam algorithm to update the second weight information and the second bias information of the long short-term memory network model; Performing supervised training on the long short-term memory network model using the processed training set to obtain a first prediction result, wherein the first prediction result is a first lowest frequency prediction value of the processed training set; Adjusting the hyperparameters of the long short-term memory network model according to the first prediction result; and, using the processed test set to verify the trained long short-term memory network model to obtain a second prediction result, wherein the second prediction result is a second lowest frequency prediction value of the processed test set; Performing a denormalization process on the second prediction result to obtain a third lowest frequency prediction value; Calculating the lowest frequency error of the long short-term memory network model according to the second lowest frequency actual value and the third lowest frequency predicted value; Determining whether the minimum frequency error is within a preset threshold value; If yes, it means that the long short-term memory network model training is completed; If not, the hyperparameters of the long short-term memory network model are updated, and the third lowest frequency prediction value is updated synchronously, and the lowest frequency error is recalculated until the lowest frequency error is within the range of a preset threshold.

8. The method for predicting the lowest frequency of disturbances in a power system based on a physical information embedded layer according to claim 1, characterized in that: The system operation data includes each bus voltage and phase angle in the power system, each generator frequency, mechanical power and electromagnetic power, load active power and system power shortage.

9. The method for predicting the lowest frequency of disturbances in a power system based on a physical information embedded layer according to claim 1, characterized in that: The simulation model is PSD-BPA.

10. The method for predicting the lowest frequency of disturbances in a power system based on a physical information embedded layer according to claim 1, characterized in that: The power system is an extended IEEE-39 node system.

Images