A method for establishing a prediction model for power system extreme cut-off time based on incremental updating
By constructing a power system extreme removal time prediction model based on incremental updating and using the dichotomy method and incremental learning training model, the solution space of the extreme removal time is simplified, the problem of slow online solution of the power system extreme removal time is solved, and an efficient extreme removal time solution is achieved.
Patent Information
- Application Number
- CN202410948236.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-15
- Publication Date
- 2025-09-05
- Estimated Expiration
- 2044-07-15
AI Technical Summary
In the existing technology, the online solution of the power system limit cut-off time is slow, especially the simulation of the data-driven method is time-consuming, which seriously limits the research efficiency.
A power system extreme cut-off time prediction model based on incremental updating is adopted. By constructing an initial prediction model and using the dichotomy method and incremental learning to train the model, the solution space of the extreme cut-off time is simplified and the algorithm solution efficiency is improved.
The solution range of the binary search method is simplified, the number of solutions is reduced, the efficiency of sample label generation and model training is improved, and the solution efficiency of the extreme resection time is increased by 90%.
Smart Images
Figure CN118944054B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of power system stability analysis, and more specifically, relates to a method for establishing a power system limit cut-off time prediction model based on incremental updating. Background Art
[0002] With the continuous expansion of my country's power grid and the increasing integration of high-proportion renewable energy and power electronic equipment, power system safety and stability analysis has become increasingly important. Preventive control is one of the three lines of defense for power systems. Among them, the limit tripping time, as a key evaluation indicator of the power system's dynamic safety and stability margin, can provide effective guidance on whether to implement preventive control measures.
[0003] Currently, there are two online methods for solving the critical resection time: direct methods and data-driven methods. With the rapid development of artificial intelligence, the speed of online solutions for critical resection time using direct methods has fallen far behind that of data-driven methods. Data-driven methods require massive amounts of data for offline training, which is based on power system simulation data and critical resection time labels. To ensure the accuracy of the critical resection time labels, time-domain simulation is generally used to continuously approximate the true value through a binary search method. However, the binary search method requires many attempts, resulting in long simulation times and severely limiting research efficiency. Therefore, simplifying the binary search solution space and improving the critical resection time are of great research significance. Summary of the Invention
[0004] In response to the defects of the existing technology and the need for improvement, the present invention provides a method for establishing a power system limit cut-off time prediction model based on incremental updating, which aims to simplify the solution space of the dichotomy method in the limit cut-off time and improve the algorithm solution efficiency.
[0005] To achieve the above-mentioned purpose, according to one aspect of the present invention, a method for establishing a power system extreme cut-off time prediction model based on incremental updating is provided, including: S1, constructing an initial prediction model, sampling and obtaining multiple batches of power system operating state quantities; S2, using a binary search method to solve the extreme cut-off time corresponding to the first batch of power system operating state quantities; S3, using the i-th batch of power system operating state quantities as input and the corresponding extreme cut-off time as a label to train the latest prediction model, where the initial value of i is 1; S4, using the trained prediction model to determine the simplified solution interval in which the extreme cut-off time corresponding to the i+1-th batch of power system operating state quantities is located, using a binary search method to determine the corresponding extreme cut-off time from the simplified solution interval, and adding one to i; S5, repeatedly executing S3-S4 until the loss function of the prediction model converges, or until the prediction model has been trained using all batches of power system operating state quantities.
[0006] Furthermore, the S3 specifically includes: when i=1, the first batch of power system operating status quantities are used as input, and the corresponding limit cut-off time is used as a label, and the initial prediction model is trained in a supervised manner; when i>1, the i-th batch of power system operating status quantities are used as input, and the corresponding limit cut-off time is used as a label, and the latest prediction model is trained through incremental learning.
[0007] Furthermore, when i=1, the method of training the initial prediction model is:
[0008]
[0009] in, is the first objective function, θ is the parameter of the prediction model, N is the number of power system operating state quantities in the same batch, represents the training set data, x (n) 、y (n) are the nth power system operating state quantity and the corresponding limit removal time in the same batch, f() represents the prediction result of the prediction model, l() represents the label loss, and Ω(θ) is the regularization term of the prediction model.
[0010] Furthermore, when i>1, the method of training the latest prediction model through incremental learning is as follows:
[0011]
[0012] Among them, F is the latest prediction model before training, θ F is the parameter of F, The parameters that need to be updated for F in this batch, is the second objective function, Represents the training set data of this batch.
[0013] Furthermore, the simplified solution interval determined by S4 is:
[0014]
[0015] in, The limit cut-off time prediction result is obtained by inputting the i+1th batch of power system operating state variables into the trained prediction model; ε max_i for The maximum absolute error between the actual limit cut-off time corresponding to the i-th batch of power system operating state quantities.
[0016] Furthermore, the S2 specifically includes: S21, determining the upper and lower limits of the limit cut-off time solution interval corresponding to the first batch of power system operating state quantities; S22, setting the fault cut-off time to the interval mean, and simulating to determine whether the power system is stable at this time. If stable, the lower limit of the limit cut-off time solution interval is set to the interval mean; if unstable, the upper limit of the limit cut-off time solution interval is set to the interval mean; S23, repeatedly executing the S22 until the interval length is less than the preset limit cut-off time accuracy.
[0017] Furthermore, the power system operating state quantities include: the amplitude and phase angle of the power system node voltage, the active power, reactive power and mechanical power of the generator, the active power and reactive power of the load, and the active transmission power and reactive transmission power between lines.
[0018] According to another aspect of the present invention, a method for determining the limit cut-off time of an electric power system based on incremental updating is provided, comprising: inputting the current electric power system operating state quantity into the prediction model obtained by the method for establishing the limit cut-off time prediction model of the electric power system based on incremental updating as described above, and the prediction model outputs the corresponding limit cut-off time.
[0019] According to another aspect of the present invention, an electronic device is provided, comprising: a processor; and a memory storing a computer-executable program, wherein when the program is executed by the processor, the processor executes the method described above.
[0020] According to another aspect of the present invention, a computer-readable storage medium is provided, on which a computer program is stored. When the program is executed by a processor, the method described above is implemented.
[0021] In general, the above technical solutions conceived by the present invention can achieve the following beneficial effects:
[0022] (1) A method for establishing a prediction model for the power system extreme cut-off time based on incremental updating is provided. The prediction model is used as an auxiliary means to guide the solution of the power system extreme cut-off time, simplifying the solution interval of the binary method, thereby reducing the number of binary method solutions, improving the efficiency of sample label generation and model training, and improving the efficiency of solving the extreme cut-off time in this stage;
[0023] Model training and sample generation are carried out in an intertwined manner. During use, the prediction model is used to improve the efficiency of sample label generation, and the obtained samples are then used to improve the effect of the prediction model, thereby improving the model training efficiency and data utilization efficiency.
[0024] (2) The simplified solution interval determined by the prediction model for the next batch of power system operating state quantities is In this interval, the binary search method is used to continue to determine the precise limit resection time. Compared with the traditional binary search method, this simplified solution interval reduces the solution calculation amount by 90%, greatly improving the training efficiency of the model. BRIEF DESCRIPTION OF THE DRAWINGS
[0025] Figure 1 A flowchart of a method for establishing a power system limit cut-off time prediction model based on incremental updating provided by an embodiment of the present invention;
[0026] Figure 2 for Figure 1 The training process of the prediction model in the method shown;
[0027] Figure 3 A single-line diagram of a New England 10-machine, 39-node system provided in an embodiment of the present invention;
[0028] Figure 4 A comparison chart between the method provided by an embodiment of the present invention and the traditional dichotomy method. DETAILED DESCRIPTION
[0029] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely for the purpose of explaining the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.
[0030] In the present invention, the terms "first", "second", etc. (if any) in the present invention and the drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.
[0031] Example 1
[0032] A method for establishing a power system limit removal time prediction model based on incremental update, see Figure 1 , combined with Figure 2-Figure 4 , the method is described in detail, and the method includes operations S1 to S5.
[0033] Operation S1: construct an initial prediction model and sample and obtain multiple batches of power system operating state quantities.
[0034] Preferably, the power system operating state quantities used in this embodiment include: the amplitude and phase angle of the power system node voltage, the active power, reactive power and mechanical power of the generator, the active power and reactive power of the load, and the active transmission power and reactive transmission power between lines.
[0035] Specifically, different generator output levels, load levels, and the proportion of motor components in the load can be set. Through power flow calculations, the amplitude and phase angle of the power system node voltage, the active power, reactive power, and mechanical power of the generators, the active power and reactive power of the loads, and the active and reactive power transmission between lines can be obtained. For example, the output of each generator can be set to fluctuate between 90% and 110% of the standard operating conditions; the load level can fluctuate between 80% and 120% of the standard operating conditions; and the proportion of motor components in the load can be set to 50%, 60%, 70%, 80%, or 90%.
[0036] Operation S2 uses a dichotomy method to solve the limit cut-off time corresponding to the first batch of power system operating state quantities.
[0037] According to an embodiment of the present invention, operation S2 specifically includes the following sub-operations S21 to S23.
[0038] In sub-operation S21 , the upper and lower limits of the limit cut-off time solution interval corresponding to the first batch of power system operating state quantities are determined.
[0039] In sub-operation S22, the fault removal time is set to the interval mean, and the simulation determines whether the power system is stable at this time. If it is stable, the lower limit of the limit removal time solution interval is set to the interval mean. If it is unstable, the upper limit of the limit removal time solution interval is set to the interval mean to form a new limit removal time solution interval.
[0040] In sub-operation S23, sub-operation S22 is repeatedly performed until the interval length is less than the preset limit excision time accuracy, and the upper limit of the interval finally obtained is used as the limit excision time.
[0041] For example, the solution interval is typically set to a lower limit of 0 seconds and an upper limit of 10 seconds for the total simulation duration. In this case, the interval average is 5 seconds. Therefore, the fault is set to clear after 5 seconds to determine the stability of the power system at this time. If the system is stable, the lower limit is changed to 5 seconds, and the solution interval is updated to 5 to 10 seconds. If the system is unstable, the upper limit is changed to 5 seconds, and the solution interval is updated to 0 to 5 seconds. This process is repeated until the interval length is less than the preset limit clearing time accuracy, such as 0.01 seconds. The upper limit of the interval can be considered the limit clearing time for the power system's operating state.
[0042] Operation S3 takes the power system operating state quantity of the i-th batch as input and the corresponding limit cut-off time as label to train the latest prediction model, and the initial value of i is 1.
[0043] When i=1, operation S3 is as follows: using the first batch of power system operating state variables as input and the corresponding limit removal time as label, the initial prediction model is trained in a supervised manner. The prediction model initialization training method is:
[0044]
[0045] in, is the first objective function, θ is the parameter of the prediction model, N is the number of power system operating state quantities in the same batch, represents the training set data, x (n) 、y (n) are the nth power system operating state quantity and the corresponding limit cut-off time in the same batch, respectively; f( ) represents the prediction result of the prediction model; l( ) represents the label loss, which is calculated by the square of the difference; Ω(θ) is the regularization term of the prediction model, which is used to avoid overfitting of the model.
[0046] When i>1, operation S3 is specifically as follows: using the power system operating state quantity of the i-th batch as input and the corresponding limit cut-off time as label, the latest prediction model is trained through incremental learning. The method of training the latest prediction model through incremental learning is:
[0047]
[0048] Among them, F is the latest prediction model before training, θ F is the parameter of F, The parameters that need to be updated for F in this batch, is the second objective function, Represents the training set data of this batch.
[0049] Operation S4, using the trained prediction model, determines the simplified solution interval of the limit cut-off time corresponding to the power system operating state quantity of the i+1th batch, uses the bisection method to determine the corresponding limit cut-off time from the simplified solution interval, and adds one to i.
[0050] Preferably, the simplified solution interval is determined as:
[0051]
[0052] in, The limit cut-off time prediction result is obtained by inputting the i+1th batch of power system operating state variables into the trained prediction model; ε max_i for The maximum absolute error between the actual limit cut-off time corresponding to the i-th batch of power system operating state quantities.
[0053] For example, in the first sample solution of three batches, the prediction results given by the prediction model after the third batch of incremental updates are During the second batch prediction process, the maximum absolute error ε between the predicted results and the true results is obtained. max_2=0.12s, then the simplified solution interval is [1.23s, 1.47s]. Finally, for this smaller interval, the bisection method is used to solve the limit removal time corresponding to the third batch of power system operating state quantities.
[0054] Operation S5 , repeatedly performing operations S3 - S4 until the loss function of the prediction model converges, or until the prediction model has been trained using all batches of power system operating state quantities.
[0055] The training process of the prediction model is as follows Figure 2 It should be noted that, in the process of training the prediction model, the limit cut-off time determined for each batch of power system operating state quantities can also be used to guide the power system cut-off and realize the determination of the limit cut-off time of the power system.
[0056] The following is an example of the New England 10-machine 39-node test system to illustrate the above method. The system single-line diagram is as follows: Figure 3 As shown in the figure, during the sample set generation phase, the output of each generator was set to fluctuate between 90% and 110% of the standard operating conditions; the load level fluctuated between 80% and 120% of the standard operating conditions; and the motor component of the load was set at 50%, 60%, 70%, 80%, and 90%, resulting in a total of 4096 power system operating states. Power flow calculations were performed using the PSASP software to obtain the required power system node voltage amplitudes and phase angles; the active power, reactive power, and mechanical power of the generators; the active power and reactive power of the loads; and the active and reactive power transferred between lines.
[0057] The power system operating status is randomly shuffled and divided into 32 batches. For all power system operating statuses in the first batch, the limit removal time is solved using the dichotomy method as the label for initial model training. Then, the power system operating status of the first batch is used as a sample, and the solved limit removal time is used as the label to train the initial auxiliary prediction model. Finally, the auxiliary prediction model is used to simplify the solution interval of the limit removal time of the next batch of samples to improve the solution efficiency. The final time result is as follows: Figure 4 As shown in the figure, it can be seen that compared with the traditional bisection method, this method can greatly reduce the simulation time and improve the efficiency of solving the limit resection time.
[0058] Example 2
[0059] A method for determining a power system's critical cutoff time based on incremental updates includes: inputting a current power system operating state into a prediction model derived from the aforementioned method for establishing a power system's critical cutoff time prediction model based on incremental updates; the prediction model then outputs a corresponding critical cutoff time. The related technical solutions are the same as those in Example 1 and are not further described here.
[0060] Example 3
[0061] An electronic device includes: a processor; and a memory storing a computer-executable program. When executed by the processor, the program causes the processor to execute the aforementioned method for establishing a power system limit cutoff time prediction model based on incremental updates, or the aforementioned method for determining a power system limit cutoff time based on incremental updates. The related technical solutions are the same as those in Example 1 and are not further described here.
[0062] Example 4
[0063] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned method for establishing a power system limit cutoff time prediction model based on incremental updates, or the aforementioned method for determining a power system limit cutoff time based on incremental updates. The related technical solutions are the same as those in Example 1 and will not be further elaborated here.
[0064] It will be easily understood by those skilled in the art that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.
Claims
1. A method for establishing a power system limit removal time prediction model based on incremental updating, characterized in that: include: S1, build the initial prediction model and sample multiple batches of power system operating status quantities; S2, using the bisection method, solves the limit removal time corresponding to the first batch of power system operating state quantities; S3, using the power system operating state quantity of the i-th batch as input and the corresponding limit removal time as label, trains the latest prediction model, and the initial value of i is 1; S4, using the trained prediction model, determining the simplified solution interval where the limit cut-off time corresponding to the power system operating state quantity of the i+1th batch lies, using a bisection method to determine the corresponding limit cut-off time from the simplified solution interval, and adding one to i; S5, repeatedly executing S3-S4 until the loss function of the prediction model converges, or until the prediction model has been trained using all batches of power system operating state quantities.
2. The method for establishing a power system limit removal time prediction model based on incremental updating according to claim 1, characterized in that: The S3 specifically includes: When i=1, the first batch of power system operating state variables are used as input and the corresponding limit removal time is used as the label to train the initial prediction model in a supervised manner; When i>1, the power system operating state quantity of the i-th batch is used as input and the corresponding limit cut-off time is used as the label, and the latest prediction model is trained through incremental learning.
3. The method for establishing a power system limit removal time prediction model based on incremental updating according to claim 2, characterized in that: When i=1, the method of training the initial prediction model is: in, is the first objective function, θ is the parameter of the prediction model, N is the number of power system operating state quantities in the same batch, represents the training set data, x (n) 、y (n) are the nth power system operating state quantity and the corresponding limit removal time in the same batch, f() represents the prediction result of the prediction model, l() represents the label loss, and Ω(θ) is the regularization term of the prediction model.
4. The method for establishing a power system limit removal time prediction model based on incremental updating according to claim 2, characterized in that: When i>1, the way to train the latest prediction model through incremental learning is: Among them, F is the latest prediction model before training, θ F is the parameter of F, The parameters that need to be updated for F in this batch, is the second objective function, Represents the training set data of this batch.
5. The method for establishing a power system limit removal time prediction model based on incremental updating according to claim 1, characterized in that: The simplified solution interval determined by S4 is: in, The limit cut-off time prediction result is obtained by inputting the i+1th batch of power system operating state variables into the trained prediction model; ε max_i for The maximum absolute error between the actual limit cut-off time corresponding to the i-th batch of power system operating state quantities.
6. The method for establishing a power system limit removal time prediction model based on incremental updating according to any one of claims 1 to 5, characterized in that: The S2 specifically includes: S21, determining the upper and lower limits of the solution interval of the limit cut-off time corresponding to the first batch of power system operating state quantities; S22, setting the fault clearing time to the interval mean, and simulating to determine whether the power system is stable at this time. If stable, the lower limit of the interval for solving the limit clearing time is set to the interval mean; if unstable, the upper limit of the interval for solving the limit clearing time is set to the interval mean; S23, repeatedly executing S22 until the interval length is less than the preset limit excision time accuracy.
7. The method for establishing a power system limit removal time prediction model based on incremental updating according to claim 1, characterized in that: The power system operating state quantities include: the amplitude and phase angle of the power system node voltage, the active power, reactive power and mechanical power of the generator, the active power and reactive power of the load, and the active transmission power and reactive transmission power between lines.
8. A method for determining the limit removal time of a power system based on incremental update, characterized in that: include: The current power system operating state quantity is input into the prediction model obtained by the method for establishing a power system limit cut-off time prediction model based on incremental update according to any one of claims 1 to 7, and the prediction model outputs the corresponding limit cut-off time.
9. An electronic device, characterized in that: include: processor; A memory storing a computer-executable program, wherein when the program is executed by the processor, the processor is caused to perform the method according to any one of claims 1 to 8.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 8 is implemented.