A fast evaluation method for power transmission section power limit based on hybrid supervised learning

Abstract

The present invention discloses a method for rapid assessment of power limit of power section based on hybrid supervised learning, comprising: (1) collecting a large amount of power section flow and SCADA data for different operation modes of the power grid to form a power grid operation mode file; (2) simulating and analyzing different operation modes of the power grid to obtain the power limit value of the power section of the corresponding operation mode; (3) using flow state information as data features, and putting the data features and the corresponding power limit value of the power section into a sample library; (4) using multiple machine learning models for supervised learning training, and selecting the optimal model from them; (5) using the selected optimal model to perform rapid assessment of power limit of power section. The present invention can form an accurate matching relationship between the operation state of the power grid and the power limit through a large amount of simulation analysis and feature engineering of historical operation modes, training and regularly updating the machine learning model, so as to quickly and accurately assess the power limit of the power section in real time.

Description

Technical Field

[0001] The present invention belongs to the technical field of electric power systems, and in particular relates to a method for quickly evaluating power limits of power transmission sections based on hybrid supervised learning. Background Art

[0002] The planning and operation of a new power system based on new energy sources urgently need to solve a series of problems such as intermittency, dynamics and uncertainty brought by a high proportion of renewable energy. Among them, timely and accurate assessment of the power transmission limit of the transmission section of the large power grid is an important part of ensuring the safe, stable and economical operation of the power grid.

[0003] The establishment of transmission section power limit usually adopts offline method, that is, modeling is carried out for the typical operation mode of the power grid in the future (comprehensively considering factors such as load level, power generation plan, renewable energy output, equipment maintenance plan, etc.), first generating a converged power flow base state file; then, using the generated base state power flow file, batch simulation analysis is carried out under various working conditions to search for power limits that meet various constraints such as power grid voltage safety, frequency safety, transient stability, voltage stability, etc. Due to the uncertainty of the increasing proportion of high renewable energy and the high complexity and nonlinearity of the power grid, the calculation and use of transmission limits of transmission sections tend to be conservative.

[0004] In order to more accurately evaluate the transmission limit to improve the safety, stability and economy of the power grid, the more common method used by independent system operators (ISOs) abroad (especially in North America) is to establish a high-precision operation model for the power grid, use the energy management system (EMS) to generate reliable real-time operating condition flow files (5-15 minutes as a cycle), and search for real-time transmission limits. This process usually requires a lot of computing resources. However, due to the continuous increase in the scale and modeling complexity of the power grid, online safety and stability simulation considering multiple safety constraints and a complete N-1 (or Nk) fault list will consume a lot of computing time and resources, and it is difficult to meet the real-time requirements of safe operation of new power systems. Therefore, the construction of new power systems urgently needs a fast and accurate online evaluation method to achieve this goal.

[0005] Network congestion is one of the major challenges facing new power systems with a high proportion of renewable energy. Once the measured power of a transmission section exceeds its power limit, grid dispatchers need to take effective control measures in a timely manner to control the section power below the limit to ensure the safe operation of the grid. The additional costs caused by grid congestion worldwide amount to billions of dollars each year. The most direct way to improve grid transmission is to build new transmission lines and other equipment, but the investment is huge and the construction period is long. Therefore, increasing the power transmission capacity of existing transmission lines while meeting multiple safety constraints, that is, "dynamic capacity expansion", is an effective solution.

[0006] To evaluate the power limit of a transmission section, it is necessary to consider a variety of safety constraints under base state and fault conditions, including thermal stability limit, stability limit, voltage safety limit, etc. The existing technology also introduces the calculation of the power limit of a transmission section. However, for a complex large power grid with multiple transmission sections, the power limit scan considering all model details and a complete fault list requires a lot of computing resources, and it is difficult to ensure the real-time performance of the power limit calculation results. Summary of the invention

[0007] The present invention provides a method for quickly evaluating the power limit of a power transmission section based on hybrid supervised learning, which can accurately evaluate the power limit value of a power transmission section online.

[0008] A method for rapid assessment of power limit of power transmission section based on hybrid supervised learning includes the following steps:

[0009] (1) According to different operation modes of the power grid, a large amount of power grid section flow information and SCADA data are collected, and the converged quasi-steady-state operation state of the power grid is obtained through flow calculation; according to future operating conditions, the power grid operation mode file is formed by adding load forecast, renewable energy forecast, maintenance plan, and power generation plan information;

[0010] (2) Simulate and analyze different operation modes i of the power grid, comprehensively consider the thermal limit TL(i), voltage limit VL(i) and stability limit SL(i), and obtain the transmission section power limit CL(i) of the corresponding operation mode;

[0011] (3) Using the quasi-steady-state power flow information of the power grid (i.e., the quasi-steady-state operation state of the power grid) as data features, the data features and the corresponding transmission section power limit values ​​are placed in a sample library; the sample library is divided into a training set, a validation set, and a test set in proportion;

[0012] (4) The data features are preprocessed as the input information of the model, the power limit value of the transmission section is used as the output information of the model, and multiple machine learning models are used for supervised learning training, and the optimal model is selected;

[0013] The training set and validation set are used to debug the model hyperparameters to improve performance, and the test set is used to evaluate the prediction performance of the model after parameter adjustment in an unknown sample set.

[0014] (5) Using the selected optimal model, a rapid assessment of the power limit of the transmission section is performed.

[0015] Furthermore, in step (2), the calculation formula of the transmission section power limit value CL(i) is:

[0016] CL(i)=min(TL(i),VL(i),SL(i)).

[0017] In step (3), the data features include: the generator active power P gen , generator reactive power Qgen, line active power Pline, line reactive power Qline, load active power Pload, load reactive power Qload, bus voltage amplitude Vm and bus voltage phase angle Va.

[0018] In step (3), the sample library is randomly split into training set, validation set and test set in a ratio of 8:1:1.

[0019] In step (4), the specific process of preprocessing the data features is: using normalization to process different types of data features, and finally forming a one-dimensional vector as the input information of the machine learning model.

[0020] Multiple machine learning models including deep neural networks, random forests, support vector regression, and gradient boosted decision trees.

[0021] In step (5), during the model application process, newly collected samples are regularly added to the sample library for continuous iterative training to improve the model accuracy and long-term effectiveness.

[0022] In step (4), one or more optimal models are selected. If multiple optimal models are selected, then in step (5), a weighted average is performed on the multiple optimal models to improve the accuracy of the overall power limit assessment. The formula is:

[0023]

[0024] Where i represents the i-th time point; CL_final represents the transmission limit assessment value of real-time dispatch operation; M is the total number of selected models; θ is a constant coefficient and satisfies CL(i,m) is the power limit value of the mth model at time i.

[0025] Compared with the prior art, the present invention has the following beneficial effects:

[0026] 1 The method of the present invention can effectively train the machine learning model and accurately evaluate the power limit value of the transmission section online.

[0027] 2. The present invention can ensure the long-term effectiveness of the method by automatically adjusting the hyperparameters to train and regularly predict the model. The effectiveness of the method is verified in a 500-node transmission network model with real operating characteristics. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 is a flow chart of an embodiment of the present invention;

[0029] Figure 2It is a single-line diagram of a power transmission section in an embodiment of the present invention;

[0030] Figure 3 is the power limit of the transmission section in the embodiment of the present invention (topology A);

[0031] Figure 4 is the power limit of the transmission section in the embodiment of the present invention (topology B);

[0032] Figure 5 The power limit prediction performance of the transmission section in the embodiment of the present invention (topology A);

[0033] Figure 6 This is the transmission section power limit prediction performance (topology B) in an embodiment of the present invention. DETAILED DESCRIPTION

[0034] The present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be pointed out that the embodiments described below are intended to facilitate the understanding of the present invention and do not have any limiting effect on the present invention.

[0035] like Figure 1 As shown in the figure, a method for rapid assessment of power limit of power transmission section based on hybrid supervised learning is proposed. The supervised learning algorithm is used to train AI agents, which can perform online assessment of power limit of power transmission section. The main process includes the following steps:

[0036] (1) Operation mode generation

[0037] According to different operation modes of the power grid, a large amount (several months or years) of power grid section flow and SCADA data is collected, and the converged power grid operation status is obtained through flow calculation; for future operating conditions, a power grid operation mode file with typical operation characteristics can be formed by adding load forecast, renewable energy forecast, maintenance plan, power generation plan and other information.

[0038] (2) Calculation of power limit of transmission section

[0039] use Figure 1 The output of the method flow and operation mode generation module shown in the figure is used to simulate and analyze different operation modes i of the power grid, comprehensively consider the thermal limit TL(i), voltage limit VL(i) and stability limit SL(i), and obtain the power transmission limit value CL(i) of the transmission section of the corresponding operation mode. The formula is:

[0040] CL(i)=min(TL(i),VL(i),SL(i))

[0041] (3) Sample library generation

[0042] A large amount of simulation data can be used to form an effective sample library to train the intelligent agent. The method proposed in this paper uses power flow information as data features, including: generator active power Pgen, generator reactive power Qgen, line active power Pline, line reactive power Qline, load active power Pload, load reactive power Qload, bus voltage amplitude Vm and bus voltage phase angle Va. For each training sample record, normalization is used to process different types of data features, and finally a one-dimensional vector is formed as the input information of the machine learning model. The output information of the model is defined as the power limit value of the selected transmission section. The sample library is generated as follows Figure 3 As shown, Ngen represents the total number of generators, Nload represents the total number of loads, Nline represents the total number of lines, and Nbus represents the total number of busbars.

[0043] (4) Model training

[0044] In order to achieve the purpose of online accurate assessment of the power limit of the transmission section, this paper uses a variety of supervised learning algorithms to train the model and select the optimal model for online application. First, the sample library is randomly split into a training set (for example, 80%), a validation set (for example, 10%), and a test set (for example, 10%) according to a certain ratio; the training set and the validation set are used to debug the model hyperparameters to improve performance.

[0045] The algorithms used in this paper include: deep neural network, random forest, support vector regression, gradient boosting decision tree, etc. In the process of model training, this paper needs to adjust the hyperparameters, that is, use multi-core parallel computing resources to search the best performing model in the validation set in the hyperparameter space to improve the overall accuracy and reliability of the model. The final output of this module is to select the best performing algorithm among different algorithm models.

[0046] (5) Model application and update

[0047] Use the test set generated by the above process to evaluate the predictive performance of the machine learning model after parameter adjustment in the unknown sample set. Finally, the model needs to be retrained using the searched optimal hyperparameters and all available samples for real-time application. It is worth noting that in order to ensure the long-term effectiveness of this method, newly collected samples need to be regularly added to the sample library to continuously iterate and improve the model accuracy and long-term effectiveness. At the same time, a variety of excellent performing algorithms can be used for weighted averaging to improve the accuracy of the overall power limit assessment. The formula is:

[0048]

[0049] Where i represents the i-th time point; CL_final represents the transmission limit assessment value used for real-time dispatch operation; M is the total number of selected models; θ is a constant coefficient and satisfies CL(i,m) is the power limit value of the mth model at time i.

[0050] In the method of the present invention, the power grid model and operation data can be obtained by modeling and simulation means or by collecting the actual operation mode of the power grid, and saved in the form of a text file; the power limit calculation of the transmission section under various safety constraints can be obtained through commercial simulation software for the power system; and the collection of sample libraries, the construction of machine learning model frameworks, model training and hyperparameter adjustment are implemented through function libraries such as Python 3.7, Tensorflow and Keras.

[0051] In order to fully test the effectiveness of the present invention, the present invention uses a 500-node system base state model and typical grid operation characteristics to generate 8,000 operating conditions, representing 8,000 hours of continuous grid operation. Table 1 below gives the power equipment statistical information of the system model.

[0052] Table 1

[0053] type describe Busbar 563 dynamo 140, with an installed capacity of 35.739GW line 571 transformer 89 load 442 Voltage level 24-500kV

[0054] In the process of sample generation, the system load changes, power generation plans and other information are first added to the grid base state model, and the power flow is calculated to obtain a converged solution. If problems such as line overload and voltage over-limit occur, the active power, reactive power and voltage control equipment of the corresponding generator set are adjusted to ensure that the generated samples meet safety requirements. Figure 2 Taking the middle transmission section as an example, the three units connected to bus 547 transmit power to the right area. The transmission section consists of five lines, including lines 547-548, 547-559, 547-549, 544-547, and 545-547.

[0055] Due to the particularity of its network structure, the power limit of this transmission section is subject to transient stability constraints. By simulating the power limit under different topological operation modes, the power limit value under various safety constraints can be obtained, such as Figure 3 and Figure 4 As shown. Figure 1 The method shown is used to form a sample library and split it into training set, validation set and test set.

[0056] For topological structures A and B under different post-fault conditions of the power grid, a variety of machine learning algorithms are used to train the transmission section power limit prediction model. In the two examples used in this paper, models such as gradient boosting decision tree (GBRT), support vector regression (SVR), deep neural network (DNN) and random forest (Random Forest) were trained respectively. The performance comparison of the models after hyperparameter adjustment is as follows: Figure 5 and Figure 6As shown, the corresponding root mean square error (RMSE), mean absolute error (MAE) and mean percentage absolute error (MAPE) are given in Tables 2 and 3 below.

[0057] Table 2 Error comparison (topology A)

[0058] Error indicators GBRT SVR DNN Random Forest RMSE(MW) 133.1 192.3 202.1 129.4 MAE(MW) 46.3 88.7 118.2 52.5 MAPE(%) 1.5657 2.9076 4.005 1.7995

[0059] Table 3 Error comparison (topology B)

[0060] Error indicators GBRT SVR DNN Random Forest RMSE(MW) 94.6 230.2 143.4 100.1 MAE(MW) 32.4 111.5 94.2 36.5 MAPE(%) 0.9768 3.1055 3.0035 1.0968

[0061] Under topology A, the MAPE prediction errors of the gradient boosting decision tree and random forest algorithms in the test set are only 1.5657% and 1.7995%; while under topology B, the MAPE prediction errors of the gradient boosting decision tree and random forest algorithms in the test set are 0.9768% and 1.0968%.

[0062] It can be seen from the comparison of test results that the method of the present invention can effectively train the supervised learning agent model and accurately evaluate the power limit value of the transmission section online, among which the gradient boosting decision tree and random forest models have the best effect. In order to improve the overall prediction accuracy of the method, the gradient boosting decision tree and random forest algorithms can be used at the same time, and the corresponding results can be weighted averaged for real-time applications.

[0063] The embodiments described above provide a detailed description of the technical solutions and beneficial effects of the present invention. It should be understood that the above are only specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, supplements and equivalent substitutions made within the scope of the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for rapid assessment of power limit of power transmission section based on hybrid supervised learning, characterized in that: The following steps are involved: (1) According to different operation modes of the power grid, a large amount of power grid section flow information and SCADA data are collected, and the converged quasi-steady-state operation state of the power grid is obtained through power flow calculation; according to future operating conditions, a file of the power grid operation mode is formed by adding load forecast, renewable energy forecast, maintenance plan, and power generation plan information; (2) Simulate and analyze different operation modes i of the power grid, comprehensively consider the thermal limit TL(i), voltage limit VL(i) and stability limit SL(i), and obtain the transmission section power limit CL(i) of the corresponding operation mode; (3) Using the quasi-steady-state operation state of the power grid as the data feature, the data feature and the corresponding transmission section power limit value are placed in the sample library; the sample library is divided into a training set, a validation set, and a test set in proportion; (4) The data features are preprocessed as the input information of the model, the power limit value of the transmission section is used as the output information of the model, and multiple machine learning models are used for supervised learning training, and the optimal model is selected; The training set and validation set are used to debug the model hyperparameters to improve performance, and the test set is used to evaluate the prediction performance of the model after parameter adjustment in an unknown sample set. (5) Using the selected optimal model, a rapid assessment of the power limit of the transmission section is performed.

2. The method for rapid assessment of power limit of power transmission section based on hybrid supervised learning according to claim 1 is characterized in that: In step (2), the calculation formula of the transmission section power limit value CL(i) is: CL(i)=min(TL(i),VL(i),SL(i)).

3. The method for rapid assessment of power limit of power transmission section based on hybrid supervised learning according to claim 1 is characterized in that: In step (3), the data features include: generator active power Pgen, generator reactive power Qgen, line active power Pline, line reactive power Qline, load active power Pload, load reactive power Qload, bus voltage amplitude Vm and bus voltage phase angle Va.

4. The method for rapid assessment of power limit of power transmission section based on hybrid supervised learning according to claim 1 is characterized in that: In step (3), the sample library is randomly split into training set, validation set and test set in a ratio of 8:1:

1.

5. The method for rapid assessment of power limit of power transmission section based on hybrid supervised learning according to claim 1, characterized in that: In step (4), the specific process of preprocessing the data features is: using normalization to process different types of data features, and finally forming a one-dimensional vector as the input information of the machine learning model.

6. The method for rapid assessment of power limit of power transmission section based on hybrid supervised learning according to claim 1, characterized in that: In step (4), multiple machine learning models include deep neural networks, random forests, support vector regression, and gradient boosted decision trees.

7. The method for rapid assessment of power limit of power transmission section based on hybrid supervised learning according to claim 1, characterized in that: In step (5), during the model application process, newly collected samples are regularly added to the sample library for continuous iterative training to improve the model accuracy and long-term effectiveness.

8. The method for rapid assessment of power limit of power transmission section based on hybrid supervised learning according to claim 1, characterized in that: In step (4), one or more optimal models are selected. If multiple optimal models are selected, then in step (5), a weighted average is performed on the multiple optimal models to improve the accuracy of the overall power limit assessment. The formula is: Where i represents the i-th time point; CL_final represents the transmission limit assessment value of real-time dispatch operation; M is the total number of selected models; θ is a constant coefficient and satisfies CL(i,m) is the power limit value of the mth model at time i.

Images