Transient frequency response curve cyclic prediction method, system and device based on long short-term memory network and medium

Abstract

The invention relates to the technical field of power transient frequency prediction, and discloses a transient frequency response curve cyclic prediction method, system and device based on a long short-term memory network, and a medium, and the method comprises the steps: obtaining a historical data set of each power generation cluster grid-connected point of a power grid, and obtaining a time sequence input characteristic quantity; different multi-layer long and short-term memory network prediction models are constructed respectively; establishing a long-time scale multilayer long-short-term memory network cycle prediction model predicted by transient characteristic quantity data through a cycle prediction method, and taking the model with the maximum prediction precision as a transient frequency response curve cycle prediction model; and measuring a time sequence input characteristic quantity of each power generation cluster grid-connected point in real time, and predicting a transient frequency response curve of each power generation cluster grid-connected point in real time after active disturbance occurs. According to the method, the spatial and temporal distribution characteristics of transient frequency response are considered, the rapid prediction method of the frequency response curve is established, and a reference basis is provided for follow-up evaluation of frequency stability.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power transient frequency prediction, and in particular to a transient frequency response curve cycle prediction method, system, device and medium based on a long short-term memory network. BACKGROUND

[0002] In recent years, the penetration rate of strong random new energy such as wind power and photovoltaic has been increasing, which has greatly reduced the inertia support capability and frequency regulation capability of the new power system. At the same time, under the background of large-scale DC grid connection, DC blocking, new energy cluster off-grid and other fault scenarios also bring possible large active power impact to the power grid, which significantly increases the risk of system frequency instability. Therefore, it is urgent to establish a more efficient and accurate transient frequency prediction model to guide the online judgment of subsequent frequency instability risk and ensure the safe operation of the system.

[0003] Compared with the traditional power grid, the new power system with the characteristics of high proportion of renewable energy and high proportion of power electronic equipment has more complex frequency stability connotation, and the traditional physical model analysis method cannot fully meet the accuracy and real-time requirements of transient frequency prediction. In recent years, with the rapid popularization of wide area measurement technology (WAMS) in the power grid, the dispatching center at all levels can quickly perceive the massive transient and steady-state operation information of the system, which provides a data premise for the application of machine learning in the power system. The powerful data processing and complex nonlinear relationship mining capability of machine learning creates new possibilities for rapid assessment of system frequency stability risk.

[0004] The power system frequency stability evaluation model based on machine learning usually takes the power grid operation state quantity obtained by the phasor measurement unit (PMU) and the SCADA (Supervisory Control And Data Acquisition) system in real time as the input feature, and uses various deep learning algorithms to fit the complex mapping relationship between the transient frequency and the input information. The existing frequency instability risk evaluation after active disturbance is not accurate. SUMMARY

[0005] In view of the above existing problems, the present application is proposed. Therefore, the present application provides a transient frequency response curve cycle prediction method based on a long short-term memory network to solve the above problems.

[0006] To solve the above technical problems, the present application provides the following technical solutions:

[0007] In a first aspect, the present application provides a transient frequency response curve cycle prediction method based on a long short-term memory network, comprising:

[0008] obtain a historical data set of each power grid cluster and grid connection point, normalize the historical data set according to categories to obtain time sequence input characteristic quantities;

[0009] extract the time sequence input characteristic quantities according to a first preset condition as a plurality of time sequence input data, and take time sequence characteristic quantities according to a second preset condition as the same as the time sequence input data as a plurality of time sequence output data, and construct different multi-layer long short-term memory network prediction models respectively;

[0010] based on the multi-layer long short-term memory network prediction model, a long time scale multi-layer long short-term memory network cycle prediction model for transient characteristic quantity data prediction is established by a cycle prediction method, and the prediction accuracy of the long time scale multi-layer long short-term memory network cycle prediction model is compared and verified, and the model with the highest prediction accuracy is taken as a transient frequency response curve cycle prediction model;

[0011] real-time measurement of time sequence input characteristic quantities of each power grid cluster and grid connection point, real-time prediction of transient frequency response curve of each power grid cluster and grid connection point after active disturbance occurs based on the transient frequency response curve cycle prediction model.

[0012] As a preferred scheme of the transient frequency response curve cycle prediction method based on the long short-term memory network, wherein: obtaining time sequence input characteristic quantities comprises:

[0013] based on the historical data set, the transient voltage vector, the active vector and the transient frequency of each power grid cluster and grid connection point of the power grid system are normalized according to categories, and are taken as time sequence input characteristic quantities of the prediction model;

[0014] The beneficial effect of the preferred scheme is to construct characteristic quantities in time sequence, which lays a high-quality, standardized data foundation for capturing dynamic dependence of data in time.

[0015] As a preferred scheme of the transient frequency response curve cycle prediction method based on the long short-term memory network, wherein: the time sequence input characteristic quantities comprise:

[0016] obtain a plurality of characteristic quantities of voltage amplitude, voltage phase angle, transient active and reactive power and transient frequency of each synchronous power cluster and wind power cluster at each sampling time, compose time sequence input characteristic quantities, and represent as:

[0017]

[0018] wherein, X n (n=1, 2, …, T) represents a characteristic vector at the nth time after active disturbance, V n , Pn, Qn, Δfn respectively represent five characteristic quantities of voltage amplitude, voltage phase angle, transient active and reactive power and transient frequency at the n th moment after the active disturbance occurs, and T is the time sequence length of the input data.

[0020] As a preferred scheme of the transient frequency response curve cycle prediction method based on the long short-term memory network, wherein: the different multi-layer long short-term memory network prediction models are respectively constructed.

[0021] The first preset condition is the time sequence characteristic quantity with a time length of 0.1s, 0.2s, 0.3s, …, 1.2s after the active disturbance;

[0022] The second preset condition is the time sequence characteristic quantity with a time length of 0.1s, 0.2s, 0.3s, …, 1.2s after the active disturbance, which meets the adjacent time period and has the same time length;

[0023] The model learning parameters are adjusted, and the multi-layer long short-term memory network prediction models S1, S2, S3, …, S 12 are respectively established between the adjacent time period characteristic quantities with different input time lengths.

[0024] The beneficial effect of the preferred scheme is that different models are constructed to provide a candidate set for the next model selection, and the model structure most suitable for such data can be finally found.

[0025] As a preferred scheme of the transient frequency response curve cycle prediction method based on the long short-term memory network, wherein: the long time scale multi-layer long short-term memory network cycle prediction model is established based on the transient characteristic quantity data prediction.

[0026] Based on the multi-layer long short-term memory network prediction models S1, S2, S3, …, S 12 , the long time scale multi-layer long short-term memory network cycle prediction models L1, L2, L3, …, L 12 are established by using the cycle prediction method, and the input data is the transient characteristic quantity within 0.1s, 0.2s, 0.3s, …, 1.2s after the active disturbance occurs, and the output data is the transient characteristic quantity 25s after the active disturbance occurs.

[0027] The beneficial effect of the preferred scheme is that the trained model is more robust to error accumulation and is more suitable for long time scale prediction.

[0028] As a preferred scheme of the transient frequency response curve cycle prediction method based on the long short-term memory network, wherein: the long time scale multi-layer long short-term memory network cycle prediction model is established.

[0029] Step one: the time sequence characteristic quantity with a time length of 0.1s, 0.2s, 0.3s, …, 1.2s after the active disturbance occurs is taken as the input characteristic quantity of the first cycle prediction, and the multi-layer long short-term memory network prediction model S1, S2, S3, …, S 12 , the multi-dimensional transient characteristic quantity V n 、 P n , Q n , Δf n is predicted.

[0030] Step two: the multi-dimensional transient characteristic quantity prediction value V n 、 P n , Q n , Δf n obtained in the first cycle is taken as the input characteristic quantity of the second cycle prediction, and the multi-layer long short-term memory network prediction model S1, S2, S3, …, S 12 predicts the multi-dimensional transient characteristic quantity in 0.3-0.4s, 0.4-0.6s, 0.6-0.9s, …, 2.4s-3.6s after the active disturbance occurs.

[0031] Repeat step one and step two, and through the cycle prediction method, the long-time scale multi-layer long short-term memory network cycle prediction model L1, L2, L3, …, L 12 .

[0032] As a preferred scheme of the transient frequency response curve cycle prediction method based on the long short-term memory network, wherein: obtaining the transient frequency response curve cycle prediction model comprises:

[0033] The prediction accuracy of the 25s transient frequency prediction curve predicted by the long-time scale multi-layer long short-term memory network cycle prediction model is quantitatively analyzed in sequence by the root mean square error, and the model with the highest prediction accuracy is selected as the transient frequency response curve cycle prediction model.

[0034] In the second aspect, the application provides a transient frequency response curve cycle prediction system based on a long short-term memory network, comprising:

[0035] The acquisition module is used for acquiring the historical data set of the grid each power generation cluster and the grid connection point, normalizing the output according to the categories of the historical data set, and obtaining the time sequence input characteristic quantity.

[0036] The first model construction module is configured to extract the time sequence input characteristic quantity according to a first preset condition as a plurality of sets of time sequence input data, and to construct different multi-layer long short-term memory network prediction models according to a second preset condition and the time sequence input data as a plurality of sets of time sequence output data.

[0037] The second model construction module is configured to establish a long time scale multi-layer long short-term memory network cycle prediction model for transient characteristic quantity data prediction based on the multi-layer long short-term memory network prediction model through a cycle prediction method, and to compare and verify the prediction accuracy of the long time scale multi-layer long short-term memory network cycle prediction model, so as to take the model with the highest prediction accuracy as a transient frequency response curve cycle prediction model.

[0038] The output module is configured to measure the time sequence input characteristic quantity of each power generation cluster grid connection point in real time, and to predict the transient frequency response curve of each power generation cluster grid connection point after the occurrence of active disturbance in real time based on the transient frequency response curve cycle prediction model.

[0039] In a third aspect, the present application provides a computer device, comprising:

[0040] a memory and a processor;

[0041] The memory is configured to store computer executable instructions, and the processor is configured to execute the computer executable instructions, so as to realize the steps of the transient frequency response curve cycle prediction method based on the long short-term memory network.

[0042] In a fourth aspect, the present application provides a computer readable storage medium, which stores computer executable instructions, and the computer executable instructions are executed by a processor to realize the steps of the transient frequency response curve cycle prediction method based on the long short-term memory network.

[0043] Compared with the prior art, the present application has the following beneficial effects: on the basis of the long short-term memory network (LSTM), the present application combines the active-frequency physical mechanism analysis with the pure data driven prediction model, and establishes a long time scale physical-data fusion driven fast online prediction model of the transient frequency response curve; the PMU measured transient operation information of the system after disturbance is used to establish a pure data driven frequency curve cycle prediction model based on the LSTM, so that the long time scale frequency response curve can be quickly predicted based on a small amount of transient information. BRIEF DESCRIPTION OF DRAWINGS

[0044] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed to be used in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description only represent some of the embodiments of the present application, and all other drawings obtained by those skilled in the art without creative labor on the basis of these drawings should also belong to the protection scope of the present application.

[0045] Figure 1 The overall flowchart of the transient frequency response curve cycle prediction method based on the long short-term memory network according to an embodiment of the present application.

[0046] Figure 2 The memory unit schematic diagram of the LSTM of the transient frequency response curve cycle prediction method based on the long short-term memory network according to an embodiment of the present application.

[0047] Figure 3 The training flowchart of the LSTM prediction model of the transient frequency response curve cycle prediction method based on the long short-term memory network according to an embodiment of the present application.

[0048] Figure 4 The frequency curve cycle prediction flowchart based on the LSTM of the transient frequency response curve cycle prediction method based on the long short-term memory network according to an embodiment of the present application.

[0049] Figure 5 The comparison schematic diagram of the results of three prediction methods under the system active disturbance of the transient frequency response curve cycle prediction method based on the long short-term memory network according to an embodiment of the present application. DETAILED DESCRIPTION

[0050] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the drawings in the specification. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should belong to the protection scope of the present application.

[0051] REFERENCE Figures 1-5 For an embodiment of the present application, a transient frequency response curve cycle prediction method based on a long short-term memory network (LSTM) is provided, which comprises:

[0052] S101, obtaining a historical data set of each power generation cluster and grid connection point of a power grid, normalizing the output of the historical data set according to categories to obtain time sequence input characteristic quantities;

[0053] S102, extract the temporal input features according to the first preset condition, use them as multiple sets of temporal input data, and use the temporal features that are the same as the temporal input data according to the second preset condition as multiple sets of temporal output data, and construct different multilayer long short-term memory network prediction models respectively.

[0054] S103, Based on the prediction model of multilayer long short-term memory network, a long-term multilayer long short-term memory network cyclic prediction model is established using the cyclic prediction method to predict transient feature data. The prediction accuracy of the long-term multilayer long short-term memory network cyclic prediction model is compared and verified, and the model with the highest prediction accuracy is taken as the cyclic prediction model of transient frequency response curve.

[0055] S104 measures the time-series input characteristics of each power generation cluster's grid connection point in real time, and predicts the transient frequency response curve of each power generation cluster's grid connection point in real time after an active power disturbance occurs based on the transient frequency response curve cyclic prediction model.

[0056] In a preferred embodiment, obtaining the temporal input feature includes:

[0057] Based on historical datasets, the transient voltage vector, active power vector, and transient frequency of each power generation cluster's grid connection point in the power grid system are normalized by category and used as time-series input features for the prediction model.

[0058] In a preferred embodiment, the temporal input features include:

[0059] At each sampling time, the voltage amplitude, voltage phase angle, transient active and reactive power, and transient frequency of each synchronous power generation cluster and wind power generation cluster grid connection point are obtained to form the time-series input feature quantity, which is expressed as:

[0060]

[0061] Among them, X n (n = 1, 2, ..., T), representing the eigenvector at time n after the active disturbance, V n , Pn, Qn, and Δfn represent five characteristic quantities at the nth moment after the active disturbance occurs: voltage amplitude, voltage phase angle, transient active and reactive power, and transient frequency, respectively. T is the time sequence length of the input data.

[0063] Specifically, based on historical datasets of the power grid system, the transient voltage vectors, active power, reactive power, and transient frequencies of each power generation cluster's grid connection point are z-score normalized according to category, and then used as the time-series input feature quantity X for the prediction model. Specifically, the time-series input feature quantity is composed of multiple features collected at each sampling time, including voltage amplitude, voltage phase angle, transient active and reactive power, and transient frequency of each synchronous generator group and wind turbine generator group's grid connection point, using wide-area measurement technology. Each feature quantity at each PMU sampling time contains information on multiple grid connection points of each synchronous generator group and wind turbine generator group in the system, namely, voltage amplitude V. n =[V n,1 V n,2 …V n,M Voltage phase angle Transient active power P n =[P n,1 P n,2 …P n,M Transient reactive power Q n =[Q n,1 Q n,2 …Q n,M Transient frequency Δf n =[Δf n,1 Δf n,2 …Δf n,M ], where M represents the number of grid connection points, and finally, z-score standardization is performed on the five types of feature quantities respectively.

[0064] In a preferred embodiment, constructing different multilayer long short-term memory network prediction models includes:

[0065] The first preset condition is a time-series characteristic quantity with a duration of 0.1s, 0.2s, 0.3s, ..., 1.2s after an active disturbance;

[0066] The second preset condition is that the time series characteristic quantities with active power disturbance durations of 0.1s, 0.2s, 0.3s, ..., 1.2s are adjacent in time period and have the same duration.

[0067] By adjusting the model learning parameters, multi-layer long short-term memory network prediction models S1, S2, S3, ..., S1 are established to predict the features between adjacent time periods with different input durations. 12 .

[0068] In a preferred embodiment, establishing a long-term scale multilayer long short-term memory network recurrent prediction model based on transient feature data includes:

[0069] Multilayer long short-term memory network prediction models S1, S2, S3, ..., S based on the features of adjacent time periods with different input durations.12 Using the cyclic prediction method, a long-scale multilayer long short-term memory network cyclic prediction model L1, L2, L3, ..., L... is established, with input data consisting of transient characteristics within 0.1s, 0.2s, 0.3s, ..., 1.2s after an active power disturbance, and output data consisting of transient characteristics 25s after the active power disturbance. 12 .

[0070] Specifically, multi-layer LSTM prediction models S1, S2, S3, ..., S based on the features of adjacent time periods with different input durations are used. 12 Using the concept of cyclic prediction, a long-scale multi-layer LSTM cyclic prediction model L1, L2, L3, ..., L4 is established, with input data consisting of transient characteristics within 0.1s, 0.2s, 0.3s, ..., 1.2s after the active power disturbance occurs, and output data consisting of transient characteristics within 25s after the active power disturbance occurs. 12 The steps are as follows:

[0071] Step 1: Using the time-series features occurring 0.1s, 0.2s, 0.3s, ..., 1.2s after the active disturbance as input features for the first cyclic prediction, the features are predicted using a multi-layer long short-term memory network between features from adjacent time periods, model S1, S2, S3, ..., S... 12 Predict the multidimensional transient characteristic V within 0.1–0.2 s, 0.2–0.4 s, 0.3–0.6 s, ..., 1.2 s–2.4 s after the occurrence of an active disturbance. n , P n Q n , Δf n .

[0072] Step 2: Calculate the predicted value V of the multidimensional transient feature obtained in the first loop. n , P n Q n , Δf n As the input features for the second iteration of prediction, the prediction model S1, S2, S3, ..., S4 is formed by a multi-layer long short-term memory network between features from adjacent time periods. 12 Predict the multidimensional transient characteristics within 0.3–0.4 s, 0.4–0.6 s, 0.6–0.9 s, ..., 2.4 s–3.6 s after the occurrence of an active disturbance;

[0073] Repeat steps one and two, and using the above cyclic prediction method, establish a long-scale multilayer long short-term memory network cyclic prediction model L1, L2, L3, ..., L5 for the transient characteristics within 25 seconds after the active disturbance occurs. 12 .

[0074] In a preferred embodiment, obtaining the transient frequency response curve cyclic prediction model includes:

[0075] By using the root mean square error, the prediction accuracy of the 25s transient frequency prediction curves predicted by the long-term multilayer long short-term memory network cyclic prediction model under the same test set was quantitatively analyzed, and the model with the highest prediction accuracy was selected as the cyclic prediction model for the transient frequency response curve.

[0076] Specifically, the root mean square error (RMSE) is used to sequentially evaluate the long-term scale recurrent prediction models L1, L2, L3, ..., L on the same test set. 12 The prediction accuracy of the predicted 25s transient frequency curve is quantitatively analyzed and expressed as follows:

[0077]

[0078] Among them, y pre Let y be the predicted curve, y be the actual curve, and t be the actual curve. j Let j be the discrete prediction time point that makes up the curve, j = 1, 2, ..., z, and z be the evaluation time period. The model with the highest prediction accuracy is selected as the online prediction model.

[0079] Finally, during online prediction, wide-area measurement technology measures the time-series input characteristic quantity X of each power generation cluster grid connection point in real time and uploads it to the dispatch center. Then, based on the transient frequency response curve cyclic prediction model, the transient frequency response curve of each power generation cluster grid connection point within 25 seconds after the active power disturbance occurs is predicted in real time.

[0080] It should be noted that, based on the Long Short-Term Memory (LSTM) network, this invention integrates active-frequency physical mechanism analysis with a pure data-driven prediction model to establish a fast online prediction model for the transient frequency response curve driven by long-term physical-data fusion. By utilizing the transient operating information of the system after measured perturbation by the PMU, a pure data-driven cyclic prediction model for the frequency curve based on LSTM is established, which can achieve rapid prediction of the long-term frequency response curve with a small amount of transient information.

[0081] The above is a schematic scheme of the transient frequency response curve cyclic prediction method based on long short-term memory networks in this embodiment. It should be noted that the technical solution of this transient frequency response curve cyclic prediction system based on long short-term memory networks belongs to the same concept as the technical solution of the transient frequency response curve cyclic prediction method based on long short-term memory networks described above. Details not described in detail in the technical solution of the transient frequency response curve cyclic prediction system based on long short-term memory networks in this embodiment can be found in the description of the technical solution of the transient frequency response curve cyclic prediction method based on long short-term memory networks described above.

[0082] This embodiment provides a transient frequency response curve cyclic prediction system based on a long short-term memory network, including:

[0083] The acquisition module is used to acquire historical datasets of the grid connection points of each power generation cluster in the power grid, normalize the output of the historical datasets by category, and obtain time-series input feature quantities.

[0084] The first model construction module is used to extract time-series input features according to the first preset conditions, and use them as multiple sets of time-series input data. It also uses the time-series features that are the same as the time-series input data according to the second preset conditions as multiple sets of time-series output data to construct different multilayer long short-term memory network prediction models respectively.

[0085] The second model construction module is used to establish a long-term multi-level long short-term memory network (MLSN) cyclic prediction model based on the cyclic prediction method using transient feature data. It compares and verifies the prediction accuracy of the long-term multi-level long short-term memory network cyclic prediction model and selects the model with the highest prediction accuracy as the cyclic prediction model for the transient frequency response curve.

[0086] The output module is used to measure the time-series input characteristics of each power generation cluster grid connection point in real time. Based on the transient frequency response curve cyclic prediction model, it predicts the transient frequency response curve of each power generation cluster grid connection point after the occurrence of active power disturbance in real time.

[0087] This embodiment also provides a computer device suitable for cyclic prediction of transient frequency response curves based on long short-term memory networks, including:

[0088] The memory and processor are used to store computer-executable instructions and execute computer-executable instructions to implement the transient frequency response curve cyclic prediction method based on long short-term memory network proposed in the above embodiments.

[0089] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the method for cyclic prediction of transient frequency response curves based on long short-term memory networks as proposed in the above embodiments.

[0090] The storage medium proposed in this embodiment and the method for predicting transient frequency response curves based on long short-term memory networks proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0091] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0092] Referring to Table 1, a method for predicting transient frequency response curves based on long short-term memory networks is provided. To verify its beneficial effects, it is scientifically demonstrated through economic benefit calculations and simulation experiments.

[0093] This invention is applied to an improved IEEE 39-node system incorporating wind power. The system's rated frequency is 50Hz. Wind farms 1 through 5 are integrated into nodes 3, 4, 16, 17, and 18 of the original IEEE 39-node system. Each wind farm consists of multiple doubly-fed induction generators (DFIGs) with a rated capacity of 2MW. The DFIGs employ a third-order model that neglects stator transients. Each DFIG in each wind farm uses droop control to regulate frequency fluctuations. The input for the droop control of each turbine is the transient frequency of the grid connection point of the synchronous generator (No. 1). PMU (Power Management Unit) devices are deployed at the synchronous generator nodes and the wind farm grid connection nodes, with the PMU sampling time frequency set to 100Hz.

[0094] This invention applies the method of the present invention to predict the transient frequency response curve of an AC system after an active power disturbance. Following the steps of the present invention, transient operating characteristics within 0.2s to 1.2s after the active power disturbance are input. The prediction accuracy based on the LSTM cyclic prediction method, and the prediction results of the gated recurrent unit (GRU), deep belief network (DBN), and LSTM-based transient frequency curve cyclic prediction model under different input durations are compared in Table 1. Figure 5 As shown.

[0095] Table 1: Comparison of prediction performance of different methods under different input data durations

[0096]

[0097]

[0098] As shown in Table 1 and Figure 5 As shown, with the increase of the evaluation period and the duration of the time-series input data, the prediction accuracy of both methods gradually improves. At different fixed evaluation periods, the prediction accuracy of the method proposed in this invention is higher than that of the GRU prediction model, thus proving the accuracy and effectiveness of the LSTM-based recurrent prediction method proposed in this invention.

[0099] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for cyclic prediction of transient frequency response curves based on long short-term memory networks, characterized in that, include: Historical datasets of power generation clusters and grid connection points of the power grid are obtained, and the output of the historical datasets is normalized by category to obtain time-series input feature quantities. The temporal input features are extracted according to the first preset condition and used as multiple sets of temporal input data. The temporal features that are the same as the temporal input data according to the second preset condition are used as multiple sets of temporal output data, and different multilayer long short-term memory network prediction models are constructed respectively. Based on the multilayer long short-term memory network prediction model, a long-term multilayer long short-term memory network cyclic prediction model based on transient feature data is established through the cyclic prediction method. The prediction accuracy of the long-term multilayer long short-term memory network cyclic prediction model is compared and verified, and the model with the highest prediction accuracy is taken as the transient frequency response curve cyclic prediction model. The time-series input characteristics of each power generation cluster grid connection point are measured in real time. Based on the transient frequency response curve cyclic prediction model, the transient frequency response curve of each power generation cluster grid connection point is predicted in real time after the active power disturbance occurs.

2. The transient frequency response curve cyclic prediction method based on long short-term memory networks as described in claim 1, characterized in that, The temporal input features obtained include: Based on the historical dataset, the transient voltage vector, active power vector, and transient frequency of each power generation cluster grid connection point in the power grid system are normalized by category and used as time-series input features of the prediction model.

3. The transient frequency response curve cyclic prediction method based on long short-term memory networks as described in claim 2, characterized in that, Temporal input features include: At each sampling time, the voltage amplitude, voltage phase angle, transient active and reactive power, and transient frequency of each synchronous power generation cluster and wind power generation cluster grid connection point are obtained to form the time-series input feature quantity, which is expressed as: Among them, X n (n = 1, 2, ..., T), representing the eigenvector at time n after the active disturbance, V n , P n Q n , Δf n These represent five characteristic quantities at the nth moment after the active disturbance occurs: voltage amplitude, voltage phase angle, transient active and reactive power, and transient frequency, respectively. T is the time sequence length of the input data.

4. The transient frequency response curve cyclic prediction method based on long short-term memory networks as described in claim 1, characterized in that, Different multi-layer long short-term memory network prediction models were constructed, including: The first preset condition is a time-series characteristic quantity with a duration of 0.1s, 0.2s, 0.3s, ..., 1.2s after an active disturbance; The second preset condition is that the time series characteristic quantities with active power disturbance durations of 0.1s, 0.2s, 0.3s, ..., 1.2s are adjacent in time period and have the same duration. By adjusting the model learning parameters, multi-layer long short-term memory network prediction models S1, S2, S3, ..., S1 are established to predict the features between adjacent time periods with different input durations. 12 .

5. The transient frequency response curve cyclic prediction method based on long short-term memory networks as described in claim 4, characterized in that, Establishing a long-term scale multilayer long short-term memory network recurrent prediction model based on transient feature data includes: Multilayer long short-term memory network prediction models S1, S2, S3, ..., S based on the features of adjacent time periods with different input durations. 12 Using the cyclic prediction method, a long-scale multilayer long short-term memory network cyclic prediction model L1, L2, L3, ..., L... is established, with input data consisting of transient characteristics within 0.1s, 0.2s, 0.3s, ..., 1.2s after an active power disturbance, and output data consisting of transient characteristics 25s after the active power disturbance. 12 .

6. The transient frequency response curve cyclic prediction method based on long short-term memory networks as described in claim 5, characterized in that, Establishing a long-term scale multilayer long short-term memory network recurrent prediction model also includes: Step 1: Using the time-series features occurring 0.1s, 0.2s, 0.3s, ..., 1.2s after the active disturbance as input features for the first cyclic prediction, the features are predicted using a multi-layer long short-term memory network between features from adjacent time periods, model S1, S2, S3, ..., S... 12 The multidimensional transient characteristic V is predicted within 0.1–0.2 s, 0.2–0.4 s, 0.3–0.6 s, ..., 1.2 s–2.4 s after the occurrence of an active disturbance. n , P n Q n , Δf n . Step 2: Calculate the predicted value V of the multidimensional transient feature obtained in the first loop. n , P n Q n , Δf n As the input features for the second iteration of prediction, the prediction model S1, S2, S3, ..., S4 is formed by multiple layers of long short-term memory network between features in adjacent time periods. 12 Predict the multidimensional transient characteristics within 0.3–0.4 s, 0.4–0.6 s, 0.6–0.9 s, ..., 2.4 s–3.6 s after the occurrence of an active disturbance; Repeat steps one and two, and use the cyclic prediction method to establish a long-scale multilayer long short-term memory network cyclic prediction model L1, L2, L3, ..., L5 for the transient characteristics within 25 seconds after the active disturbance occurs. 12 .

7. The transient frequency response curve cyclic prediction method based on long short-term memory networks as described in claim 6, characterized in that, The cyclic prediction model for the transient frequency response curve includes: Using the root mean square error, the prediction accuracy of the 25s transient frequency prediction curves predicted by the long-term multilayer long short-term memory network cyclic prediction model under the same test set is quantitatively analyzed, and the model with the highest prediction accuracy is selected as the cyclic prediction model for the transient frequency response curve.

8. A transient frequency response curve cyclic prediction system based on long short-term memory networks, employing the transient frequency response curve cyclic prediction method based on long short-term memory networks as described in any one of claims 1 to 7, characterized in that, include, The acquisition module is used to acquire historical datasets of the grid connection points of each power generation cluster in the power grid, and to normalize the output of the historical datasets by category to obtain time-series input feature quantities. The first model construction module is used to extract the temporal input features according to the first preset condition, and use them as multiple sets of temporal input data. The temporal features that are the same as the temporal input data according to the second preset condition are used as multiple sets of temporal output data to construct different multilayer long short-term memory network prediction models respectively. The second model construction module is used to establish a long-term multi-level long short-term memory network cyclic prediction model based on the multi-level long short-term memory network prediction model and to establish a long-term multi-level long short-term memory network cyclic prediction model using transient feature data through a cyclic prediction method. The prediction accuracy of the long-term multi-level long short-term memory network cyclic prediction model is compared and verified, and the model with the highest prediction accuracy is taken as the transient frequency response curve cyclic prediction model. The output module is used to measure the time-series input characteristics of each power generation cluster grid connection point in real time, and based on the transient frequency response curve cyclic prediction model, to predict the transient frequency response curve of each power generation cluster grid connection point after the occurrence of active power disturbance in real time.

9. A computer device, characterized in that, include: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, they implement the steps of the transient frequency response curve cyclic prediction method based on long short-term memory network as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, It stores computer-executable instructions, which, when executed by a processor, implement the steps of the transient frequency response curve cyclic prediction method based on a long short-term memory network as described in any one of claims 1 to 7.

Images