Power system stabilization system and power system stabilization method

Abstract

To provide a power system stabilization system and a power system stabilization method that can reduce the maximum prediction error by artificial intelligence and speed up fault calculations.SOLUTION: A system stabilization system includes: a fault state change calculation unit that calculates a fault state change amount using a pre-fault state and a post-fault state as input; a fault calculation estimation model learning unit that takes the fault state change amount, a system model, a pre-fault state, a fault case, and a fault calculation estimation model as input, learns an initial model of the fault calculation estimation model, and generates a learned fault calculation estimation model; a fault state change estimation unit that takes the system state, the system model, the fault calculation estimation model, and a fault case as input, and calculates a fault state change amount estimated value; a fault state estimation unit that estimates a fault state from the fault state change amount estimated value and the system state; and a display unit that displays one or more of the input and output data.SELECTED DRAWING: Figure 7

Description

[Technical field]

[0001] The present invention relates to a power system stabilization system and a power system stabilization method related to the operation, planning, and protection of a power system, and is suitably applied to a power system stabilization system that uses artificial intelligence to stabilize a power system, for example. [Background technology]

[0002] In recent years, with the increase in renewable energy sources connected to the power grid and the electrification of consumer equipment, there are concerns that overloads on transmission lines that were not previously anticipated may occur.Even if an overload does occur, there are systems that can eliminate the overload by switching the output of renewable energy sources, known as special protection schemes (SPS).

[0003] In power system stabilization systems and power system operation systems, the state of operation has been evaluated and measures taken for serious conditions by confirming the state of the power system fault by simulating it in advance. It is known that the calculation process of such simulations requires particularly large calculation resources when used during operation, and various speed-up methods using screening and new computers have been considered.

[0004] In response to these challenges in the power grid field, the field of information science has been drawing attention for using artificial intelligence to substitute for various calculations. It is known that artificial intelligence is particularly effective at learning from training data and repeating learned tasks. There is a movement to use such artificial intelligence to speed up calculations.

[0005] Patent Document 1 describes a power system power flow calculation method that utilizes an artificial neural network. Non-Patent Document 1 describes that in order to speed up the power system power flow calculation method, an initial point of the power flow calculation is calculated using a neural network, and the power flow calculation is performed from this initial point. Non-Patent Document 1 describes the use of both the DC method and the AC method. Meanwhile, Non-Patent Document 2 describes the use of a graph neural network. [Prior art documents] [Patent documents]

[0006] [Patent Document 1] US Patent Publication No. 12 / 923525 [Non-patent literature]

[0007] [Non-Patent Document 1] Diehl, F. (2019). Warm-Starting AC Optimal Power Flow with Graph Neural Networks. [Non-Patent Document 2] TB Lopez-Garcia and JA Dominguez-Navarro, “Graph Neural Network Power Flow Solver for Dynamical Electrical Networks,” 2022 IEEE 21st Mediterranean Electrotechnical Conference (MELECON), Palermo, Italy, 2022, pp. 825-830, doi: 10.1109 / MELECON53508.2022.9842974. Summary of the Invention [Problem to be solved by the invention]

[0008] Patent Document 1, Non-Patent Document 1, and Non-Patent Document 2 each propose a calculation speed-up method that utilizes artificial intelligence using different methods. A common problem with these methods is that, as a result of emphasizing versatility, the range of prediction errors by artificial intelligence cannot be narrowed down. In addition, Non-Patent Document 1 describes using it as the initial point of the power system's power flow calculation function as a way to deal with this prediction error. While this is one effective means for minimizing the prediction error, if there is a large prediction error, it increases the number of calculation steps, which may not be effective in speeding up the calculation process.

[0009] The present invention has been made in consideration of the above points, and aims to propose a power system stabilization system and a power system stabilization method that can reduce the maximum prediction error by artificial intelligence and speed up calculations. [Means for solving the problem]

[0010] In order to solve such problems, the present invention includes a fault state change calculation unit that calculates a fault state change amount using a pre-fault state and a post-fault state as input, a fault calculation estimation model learning unit that inputs the fault state change amount, a system model, the pre-fault state, and a fault case as objective variables, learns an initial model of a fault calculation estimation model, and generates a learned fault calculation estimation model, a fault state change estimation unit that calculates a fault state change amount estimate using a system state, the system model, the learned fault calculation estimation model, and the fault case as input, a fault state estimation unit that estimates a fault state based on the fault state change estimate value and the system state, and a display unit that displays one or more of the input and output data.

[0011] Furthermore, in the present invention, the power system stabilization system has the following steps: a fault state change calculation step in which a fault state change calculation unit of the power system stabilization system calculates a fault state change amount by using a pre-fault state and a post-fault state as inputs; a fault calculation estimation model learning step in which a fault calculation estimation model learning unit of the power system stabilization system inputs the fault state change amount, a system model, the pre-fault state, and a fault case as objective variables, learns an initial model of a fault calculation estimation model, and generates a learned fault calculation estimation model; a fault state change estimation step in which a fault state change estimation unit of the power system stabilization system inputs a system state, the system model, the learned fault calculation estimation model, and the fault case as inputs, and calculates a fault state change amount estimated value; a fault state estimation step in which a fault state estimation unit of the power system stabilization system estimates a fault state based on the fault state change estimated value and the system state; and a display step in which a display unit of the power system stabilization system displays one or more of input / output data. Effect of the Invention

[0012] According to the present invention, it is possible to realize a power system stabilization system and a power system stabilization method that can reduce the maximum prediction error by artificial intelligence and speed up calculations. [Brief description of the drawings]

[0013] [Figure 1] FIG. 1 is a diagram illustrating an example of a power system according to an embodiment of the present invention. [Figure 2A] FIG. 1 is a diagram illustrating an example of a power system according to an embodiment of the present invention. [Figure 2B] FIG. 1 is a diagram illustrating an example of a power system according to an embodiment of the present invention. [Diagram 3] FIG. 2 is a diagram showing an example of an image of the principle of power flow calculation in this embodiment. [Figure 4] FIG. 13 is a diagram showing an example of calculation when a fault occurs in a power system. [Figure 5A] FIG. 1 is a diagram illustrating an example of a general principle of speedup; [Figure 5B] FIG. 1 is a diagram illustrating an example of the principle of speedup according to the present embodiment. [Figure 6] FIG. 2 is a diagram illustrating an example of a hardware configuration of the power grid stabilization system according to the present embodiment. [Figure 7] FIG. 2 is a diagram illustrating an example of a functional configuration of the power grid stabilization system according to the present embodiment. [Figure 8] 4 is a flowchart showing an example of a procedure of an operation example of the power system stabilization system according to the present embodiment. [Figure 9] FIG. 2 is a diagram showing an example of a learning method for the power grid stabilization system according to the present embodiment. [Figure 10] FIG. 4 is a diagram showing an example of learning of the power grid stabilization system according to the present embodiment. [Figure 11] FIG. 2 is a diagram showing an example of a display of the power grid stabilization system according to the present embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0014] An embodiment of the present invention will now be described in detail with reference to the drawings.

[0015] (1) Power system An example of a power system in this embodiment will be described. Fig. 1 is a simplified diagram illustrating an example of a connection configuration of the power system GR1. The power system GR1 includes generators G1-G3, loads LD1-LD3, buses B1-8, and transmission lines L1-L9. The purpose of the power system GR1 is to supply power from any of the generators G1-G3 to any of the loads LD1-LD3 via any of the buses B1-8 and transmission lines L1-L9.

[0016] Preventive control is realized by controlling each element in the power system to suppress the impact on the transmission lines L1 to L9 in the event of a fault. Examples of this preventive control include changing the output of the generators G1 to G3, requesting cooperation from the loads LD1 to 3 (demand response), and changing the connection configuration.

[0017] Next, types of faults that may occur in the power system GR1 will be described. FIG. 2A and FIG. 2B each show an example of a state in which an event occurs in the power system GR1. As described above, various events may occur in the power system GR1 due to a wide range of factors. For example, in the power system GR1a, if the generator G1 is a renewable energy power source, the generator G1 is disconnected from the system. In addition, in the power system GR1b, a lightning strike may cause a transmission line to open, causing a change in the system configuration.

[0018] In either case of fault, the power flow in each transmission line and the voltage at each bus bar change. If an event such as that shown in power systems GR1a to GR1b causes a dramatic change in the power flow, for example exceeding 150%, the event may cause a chain reaction.

[0019] Here, specific problems in this embodiment will be described. In the above-mentioned post-fault state, conventionally, a calculation called power flow calculation is repeated for each assumed fault to estimate. Here, power flow calculation is a calculation in which the bus voltage V and bus phase θ are found from initial information by utilizing an optimization algorithm, and active power P and reactive power Q are calculated by circuit calculation of the power system.

[0020] (2) Principles of load flow calculations in power systems Figure 3 shows an example of the concept behind the power system power flow calculation. For example, starting from the initial point where the bus voltage V and bus phase θ are set to 1 and 0, respectively, an optimal point is found that is the best evaluation function (the most consistent in terms of circuit calculation). This optimal point indicates the power flow calculation result. In this case, the search for this optimal point is generally carried out by searching for the optimal solution using a circuit calculation formula generated from the power system model.

[0021] Figure 4 shows an example of calculations when a fault occurs in power system GR1. When a fault occurs in power system GR1, the range in which the power flow equation of power system GR1 holds true changes. Therefore, it is necessary to search for the optimal point again, using the state before the fault as the initial point. This search process must be repeated for each fault location, making evaluation difficult. This is the calculation overhead when performing fault calculations, and the reason why it is difficult to speed up.

[0022] Figure 5A shows an example of using a general-purpose artificial intelligence model for power flow calculation. In a general-purpose artificial intelligence model for power flow calculation, a new optimal point is ultimately found by inputting a specified power system configuration, power generation, consumption, reactive power injection, etc. In this case, the new optimal point is not found directly, but the initial point is moved to a position near the new optimal point, and the optimal point is reached using a specified routine.

[0023] Here, in principle, if the prediction by the artificial intelligence is sufficient, it is considered that the number of steps to the optimal point will be reduced, and the calculation time to complete the calculation for each fault case will be reduced. On the other hand, if the prediction by the artificial intelligence is significantly off, the opposite effect may be obtained. This is because when the power flow calculation itself is used as a substitute, a learning method that targets the entire predetermined solution space is used, so the range of results that the artificial intelligence can calculate is wide. In this embodiment, the maximum prediction error is reduced in order to perform the fault calculation.

[0024] Here, we will briefly explain the characteristics of general fault calculations. General power flow calculations during faults can be understood as calculating the transition to the state after the fault, starting from the state before the fault. At this time, it is known that the state after the fault does not deviate significantly from the state before the fault. Therefore, if the range that the artificial intelligence should learn is limited to the difference before and after the fault, it is easy to see that large prediction errors are unlikely to occur.

[0025] Of course, the versatility of the calculation will decrease, but, for example, it is now known that the state of an entire power system can be calculated in a matter of seconds, and it may not seem necessary to speed up the load flow calculation itself. However, when there are a large number of objects to be calculated, it is inevitably desirable to speed up the calculation.

[0026] 5B shows an example of the principle of speeding up according to this embodiment. In this embodiment, by predicting the amount of change specifically for calculations at the time of failure, the prediction range of the artificial intelligence can be limited and the maximum prediction error can be controlled.

[0027] (3) Hardware Configuration 6 shows an example of a hardware configuration of the power system stabilization system 1. The power system stabilization system 1 includes a program database DB1, a database DB2, an input unit 91, a calculation unit 92, a display unit 93, a memory 94, a communication unit 95, a communication network 96, a measuring instrument 97, a control terminal 98, and a bus 99.

[0028] The program database DB1 includes a set of programs for calculation processing related to the power system stabilization system 1. The database DB2 stores the results of the calculation processing, etc. The input unit 91 includes, for example, a keyboard, a mouse, a stylus, a microphone, and the like.

[0029] The calculation unit 92 executes calculation processing and includes, for example, any one of a central processing unit (CPU), a graphic processing unit (GPU), a field programmable gate array (FPGA), and a tensor processing unit (TPU).

[0030] The display unit 93 displays input data and the results of system calculations, and is, for example, a monitor. The power system stabilization system 1 may also include a speaker.

[0031] The memory 94 stores various programs when they are executed, and stores the results of calculations performed by the calculation unit 92 .

[0032] The communication unit 95 communicates data with a measuring instrument 97 and a control terminal 98 via a communication network 96. The program database DB1, database DB2, input unit 91, calculation unit 92, display unit 93, memory 94, and communication unit 95 are connected via a bus 99 and transmit and receive data to and from each other. The display unit 93 performs display based on the data in the memory 94. The bus 99 does not necessarily have to be a physical bus of a single computer, and may be a bus that plays a similar role when multiple computers are combined.

[0033] The measuring instrument 97 is installed in the power system and measures, for example, one or more of effective values, phasor forms, and three-phase instantaneous values ​​of voltage, current, voltage phase, frequency, and current phase. Examples include SV / TM and PMU, and the measuring instrument 97 transmits the measurement data to the communication unit 95 of the power system stabilization system 1 via the communication network 96.

[0034] The control terminal 98 transmits operation commands to the communication unit 95 via the communication network 96 to change the state of the power system itself, such as a relay that can change the configuration of the power system, in response to operations by an operator.

[0035] (4) Functional configuration of the power grid stabilization system Next, an example of the functional configuration of the power system stabilization system 1 according to this embodiment will be described with reference to Fig. 7. The power system stabilization system 1 includes a pre-fault state DB3, a post-fault state DB4, a system model DB5, a fault case DB6, a fault state change amount calculation unit 2, a fault state change amount DB7, a fault calculation estimation model learning unit 3, a fault calculation estimation model DB8, a display unit 93, a system state DB9, a fault state change amount estimation unit 4, a fault state change amount estimated value DB10, a fault state estimation unit 5, a correction unit 6, an online corrected post-fault state DB11, a fault calculation countermeasure unit 7, and a control unit 8.

[0036] In the power system stabilization system 1, the fault state change amount calculation unit 2 and the fault calculation / estimation model learning unit 3 operate offline as follows.

[0037] The fault state change amount calculation unit 2 receives the pre-fault state DB3 and the post-fault state DB4 as input, and generates a fault state change amount DB7.

[0038] The fault calculation estimation model learning unit 3 inputs the fault state change amount DB7, the system model DB5, the pre-fault state DB3 and the fault case DB6, learns an initial model of the fault calculation estimation model DB8, and generates a learned fault calculation estimation model DB8.

[0039] Furthermore, in addition to the system model DB5, pre-fault state DB3, and fault case DB6, the fault calculation estimation model learning unit 3 further inputs the learned fault calculation estimation model DB8 as an explanatory variable and inputs the fault state change amount DB7 as a predictive variable, and learns the fault calculation estimation model DB8.

[0040] On the other hand, in the power system stabilization system 1, the fault state change amount estimating unit 4, the fault state estimating unit 5, the correcting unit 6, the fault calculation countermeasure unit 7, and the control unit 8 operate online as follows.

[0041] The fault state change amount estimating unit 4 receives as input the system state DB9, the system model DB5, the learned fault calculation and estimation model DB8, and the fault case DB6, and generates a fault state change amount estimated value DB10.

[0042] The fault state estimation unit 5 receives the fault state change amount estimated value DB 10 and the system state DB 9 as input, and estimates the post-fault state.

[0043] The corrector 6 generates an online corrected post-fault state DB 11 based on the system model DB 5 and the fault state estimated by the fault state estimator 5.

[0044] The fault calculation countermeasure unit 7 performs calculations to plan countermeasures for the fault calculation results using the online corrected post-fault state DB 11 as input. The control unit 8 is an example of a fault calculation unit, and executes fault countermeasures according to the contents of the fault calculation results.

[0045] The display unit 93 displays various data, for example, one or more of the following input / output data. Examples of the input / output data include a pre-fault state DB3, a post-fault state DB4, a system model DB5, a fault case DB6, a fault state change amount DB7, a fault calculation estimation model DB8, a system state DB9, a fault state change amount estimated value DB10, and a post-fault state after online correction DB11, or any combination of these.

[0046] (5) Input / Output Data The pre-fault state DB3 represents the state of the power system, and in particular, indicates the state before the occurrence of a fault in the power system. The pre-fault state DB3 includes, for example, one or more of the voltage and voltage phase of each bus, the active power and reactive power flowing in the transmission line and transformer, and the on / off state of the power transmission equipment.

[0047] The post-fault state DB4 represents the state of the power system, and in particular, the pre-fault state equipment configuration has been partially changed. The post-fault state DB4 includes, for example, one or more of the voltage and voltage phase of each bus, the active power and reactive power flowing through the transmission line and transformer, and the on / off state of the power transmission equipment.

[0048] The system model DB5 stores a model of the configuration and components of the power system, and includes, for example, the impedance of the transmission line, the position of the load, the upper and lower limits of the generator, and the like.

[0049] The fault case DB6 lists faults in the power system, and includes data items such as fault message, fault type, and fault details. Examples of data items may be stored in the form of "transmission line L1 fault", "transmission line fault", and "fault and open on transmission line L1".

[0050] The fault state change amount DB7 indicates the difference between the pre-fault state DB3 and the post-fault state DB4. The fault state change amount DB7 includes one or more of the amount of change in voltage, the amount of change in phase, the amount of change in phase difference, the amount of change in active power, and the amount of change in reactive power.

[0051] The fault calculation estimation model DB8 is a model that can obtain an output for an input, such as a neural network, and includes a function used in the model and a parameter to be learned. The structure of the fault calculation estimation model DB8 must be specified in advance, and it is the parameters that are determined by learning.

[0052] The system state DB9 is the state of the system obtained when the system stabilization system is in online operation, and the data format may be the same as the pre-fault state DB3 and the post-fault state DB4.

[0053] The fault state change amount estimated value DB10 is an estimate of the fault state change amount when a fault occurs in the system state DB 9. The fault state change amount estimated value DB10 can have the same data content as the fault state change amount DB7.

[0054] The online corrected post-fault state DB11 is an estimate of the corrected post-fault state obtained by adding together the system state DB9 and the fault state change amount estimated value DB10 by the correction unit 6, and has the same data format as the post-fault state DB4.

[0055] The fault state change amount calculation unit 2 calculates the difference from the pre-fault state DB 3 and the post-fault state DB 4. This difference plays an important role in realizing this embodiment, and is effective in reducing the range of prediction error.

[0056] The fault calculation estimation model learning unit 3 receives the pre-fault state DB3, the system model DB5, the fault case DB6, the fault state change amount DB7, and the fault calculation estimation model DB8 as inputs, and causes the fault calculation estimation model DB8 to learn.

[0057] The fault state change amount estimating unit 4 calculates a fault state change amount estimated value DB10 from the system state DB9, the system model DB5, the fault case DB6 and the fault calculation estimation model DB8.

[0058] The fault state estimation unit 5 estimates a fault state from the fault state change amount estimated value DB 10 and the system state DB 9 , and outputs the estimate to the correction unit 6 .

[0059] The correction unit 6 calculates an online corrected post-fault state DB11 using the system model DB5 and the output of the fault state estimation unit 5. The fault calculation countermeasure unit 7 calculates countermeasures for faults for unstable states of online fault states. The control unit 8 executes countermeasures for fault calculations calculated by the fault calculation countermeasure unit 7.

[0060] The display unit 93 displays the various input and output data described above.

[0061] (7) Learning and prediction 8 is a flowchart showing an example of the procedure of a power system stabilization method as an operation example of the power system stabilization system. An offline portion and an online portion will be described separately. Here, an overview of the power system stabilization method according to this embodiment will be described. The power system stabilization method includes a fault state change amount calculation step in which a fault state change amount calculation unit 2 of the power system stabilization system 1 inputs a pre-fault state DB3 and a post-fault state DB4 to calculate a fault state change amount DB7, and a fault calculation estimation model learning unit 3 of the power system stabilization system 1 inputs the fault state change amount DB7, a system model DB5, a pre-fault state DB3, and a fault case DB6, learns an initial model of the fault calculation estimation model DB8, and generates a learned fault calculation estimation model DB8. The method includes a model learning step, a fault state change estimation step in which the fault state change estimation unit 4 of the power system stabilization system 1 receives the system state DB9, the system model DB5, the learned fault calculation estimation model, and the fault case DB6 as inputs and calculates a fault state change estimate value DB10, a fault state estimation step in which the fault state change estimation unit 4 of the power system stabilization system 1 estimates a fault state based on the fault state change estimate value DB10 and the system state DB9, and a display step in which the display unit 93 of the power system stabilization system 1 displays one or more of the input and output data. First, an offline description will be given.

[0062] In step S1, the fault state change amount calculation unit 2 reads various data (for example, pre-fault state DB3 and post-fault state DB4). In step S2, the fault state change amount calculation unit 2 calculates the fault state change amount based on the difference between the pre-fault state DB3 and the post-fault state DB4. In this step S2, as described above, the fault state change amount calculation unit 2 calculates the difference between the pre-fault state DB3 and the post-fault state DB4, and this difference utilizes, for example, one or more of the difference in bus voltage phase (V and θ) and the difference between the active power P and reactive power Q flowing through each line. This makes it possible to define the amount of change in the state and extract the value to be learned.

[0063] In step S3, the fault calculation estimation model learning unit 3 generates a learned fault calculation estimation model DB8 from the initial model of the fault calculation estimation model DB8. In this step S3, the fault calculation estimation model learning unit 3 inputs the fault state change amount DB7, the system model DB5, and the fault case DB6, and learns the initial model of the fault calculation estimation model DB8.

[0064] Furthermore, in addition to the system model DB5, the pre-fault state DB3, and the fault case DB6, the fault calculation estimation model learning unit 3 further inputs the learned fault calculation estimation model DB8 as an explanatory variable and inputs the fault state change amount DB7 as a predictive variable to learn the fault calculation estimation model DB8. In this way, by dividing the input data into explanatory variables and predictive variables, preparation for learning can be made.

[0065] At this time, the fault calculation estimation model DB8 is basically a kind of regression model, and may utilize, for example, a neural network. At this time, the fault calculation estimation model DB8 as a learning model must be able to output the post-fault state, and in this embodiment, an activation function with a limited output range is utilized. This is particularly effective, and contributes to keeping the error of the output layer, which will be described later, within a predetermined range.

[0066] In principle, it is effective to make the target predicted by the fault calculation estimation model DB8 small. Note that input and output data may be normalized during learning. The structure of the fault calculation estimation model DB8 does not need to be specified in particular, but it may be configured in a form that simulates the physical configuration of the system or as a graph neural network.

[0067] In step S4, the fault calculation estimation model learning unit 3 outputs the learned fault calculation estimation model DB 8. The above is the offline preparation process, and next, the online process will be described.

[0068] First, in step S5, the fault state change amount estimation unit 4 reads, for example, the pre-fault state DB3 and the trained fault calculation estimation model DB8. In step S6, the fault state change amount estimation unit 4 inputs, for example, the pre-fault state DB3, and estimates the fault state change amount estimated value DB10 as the fault state change amount from the trained fault calculation estimation model DB8. In this step S6, it is assumed that the system will be used online, and the amount of change to the fault state is predicted for the actual system state. However, unlike the learning phase, it is not necessary to change the structure of the fault calculation estimation model DB8. By converting this change amount into a post-fault state predicted value in step S7, the previous fault calculation is partially substituted.

[0069] In step S7, the fault state estimation unit 5 estimates the post-fault state from the fault state change amount estimated value DB10 (and preferably the system state DB9).

[0070] In step S8, the correction unit 6 corrects the post-fault state using power flow calculation. In step S9, the correction unit 6 outputs the post-fault state. More specifically, in step S8, the correction unit 6 can reduce the calculation steps of the previous fault calculation until the optimal point is reached using the post-fault state predicted value as the initial point. The correction function by the correction unit 6 utilizes power flow calculation. By using such a correction function, it is possible to obtain the effect of suppressing prediction errors by setting a prediction range of the amount of change as shown in FIG. 5B, and it is possible to maintain calculation accuracy. The above is the process up to online evaluation, and the following is the process up to online control.

[0071] In step S10, the fault calculation countermeasure unit 7 performs calculations to plan countermeasures for the fault calculation result using the online corrected post-fault state DB 11 as input. In step S11, the control unit 8 executes fault countermeasures according to the contents of the fault calculation result.

[0072] In this embodiment, through such a series of procedures, the series of functions described above can be used to execute fault calculations, while also allowing corresponding countermeasures to be taken against faults.

[0073] 9 is a diagram showing an example of a learning method of the system stabilization system of this embodiment. The illustrated example shows how a pre-fault state DB3, a system model DB5, and a fault case DB6 are input as explanatory variables to a trained fault calculation estimation model learning unit 3 online to predict a fault state change amount DB7. In the fault calculation estimation model learning unit 3, a fault calculation estimation model DB8 includes an output layer DB81 that limits the output range, and since the output layer DB81 has a limit, it is effective in reducing the range of prediction errors.

[0074] Next, the learning phase will be described. Fig. 10 explains an example of the behavior of the fault calculation estimation model DB8 during learning, and is used to evaluate the validity of the learning. During learning, the data is divided into learning data that affects the internal parameters of the fault calculation estimation model DB8 and evaluation data that does not affect the internal parameters, and a series of learning is performed.

[0075] In this case, as the evaluation axis for learning, RMSE (Root Mean Square Error), maximum error, etc. may be used, and an optimization routine such as ADAM may be used. Specifically, as the evaluation axis for learning, any one of the following is used: the maximum error between the predicted value of the fault state change amount for calculating the fault calculation estimation model DB8 and the fault state change amount, the root mean square error (RMSE) between the predicted value of the fault state change amount for calculating the fault calculation estimation model DB8 and the fault state change amount, and the sum of absolute values ​​of errors between the predicted value of the fault state change amount for calculating the fault calculation estimation model DB8 and the fault state change amount.

[0076] As a result, the fault calculation estimation model learning unit 3 learns in a direction that minimizes one or more of the maximum error between the predicted value of the fault state change amount used to calculate the fault calculation estimation model DB8 and the fault state change amount, the squared sum error between the predicted value of the fault state change amount used to calculate the fault calculation estimation model DB8 and the fault state change amount, and the sum of the absolute values ​​of the error between the predicted value of the fault state change amount used to calculate the fault calculation estimation model DB8 and the fault state change amount.

[0077] The learning method can be effectively performed on either a CPU, GPU, or TPU. As shown in Figure 10, as the number of learning iterations increases, the evaluation items of both the evaluation data and the learning data move in a favorable direction (in this case, since it is an error, lower is better). If the value of the evaluation function (value of the evaluation axis) does not fall any further even if the number of learning iterations is increased, it can be seen that learning is complete.

[0078] At this time, the fault calculation estimation model learning unit 3 judges the validity of the learning result for the learned fault calculation estimation model, and may use one or more of the following as a criterion for judging the validity of the learning result: setting a threshold value for the value of the evaluation function; and evaluating the total calculation time that can be reduced by performing a simulated online calculation using the learning result. In the power flow calculation in the fault calculation targeted in this embodiment, the calculation time is reduced by changing the number of times until the power flow calculation converges, so that the online calculation time can be reduced by evaluating using data that assumes the online system state.

[0079] Here, when learning of the learned fault calculation estimation model DB8 is insufficient, the fault calculation estimation model learning unit 3 can use one or more of the following means: changing the structure of the fault calculation estimation model DB8, changing the learning parameters of the fault calculation estimation model DB8, increasing the input data of the fault calculation estimation model DB8, decreasing the input data of the fault calculation estimation model DB8, and changing the format (expression method) of the input data of the fault calculation estimation model DB8. In particular, when changing the format (expression method) of the input data, for example, a method of limiting to some parameters and expressing only the change amount thereof, instead of entering all of the system model DB5, can be used.

[0080] (8) Display examples Fig. 11 shows an example of display by the power system stabilization system 1. In this embodiment, the control unit 8 has a function of displaying one or more of the input and output data, and as an example, as shown in Fig. 11, displays the system (Line), pre-fault state (Base), change prediction of the system state (Change Estimation), and the result corrected thereafter (Final Result).

[0081] By using such a display result, it is possible to predict the amount of change in the state of the fault and then understand that the change has been corrected through calculation. This embodiment is expected to improve calculation speed, and at the same time, it is possible to recognize that the results are reliable.

[0082] The power system stabilization system 1 according to this embodiment includes a fault state change calculation unit 2 that calculates a fault state change amount DB7 using a pre-fault state DB3 and a post-fault state DB4 as input, a fault calculation estimation model learning unit 3 that inputs the fault state change amount DB7, a system model DB5, a pre-fault state DB3, and a fault case DB6, learns an initial model of a fault calculation estimation model DB8, and generates a learned fault calculation estimation model DB8, a fault state change estimation unit 4 that calculates a fault state change amount estimated value DB10 using an input of a system state DB9, the system model DB5, the learned fault calculation estimation model DB8, and a fault case DB6, a fault state estimation unit 5 that estimates a fault state based on the fault state change amount estimated value Db10 and the system state Db9, and a display unit 93 that displays one or more of the input and output data.

[0083] The power system stabilization method of the power system stabilization system 1 according to the present embodiment includes a fault state change amount calculation step in which a fault state change amount calculation unit 2 of the power system stabilization system 1 calculates a fault state change amount DB7 by inputting a pre-fault state DB3 and a post-fault state DB4, and a fault calculation estimation model learning step in which a fault calculation estimation model learning unit 3 of the power system stabilization system 1 inputs the fault state change amount DB7, a system model DB5, a pre-fault state DB3, and a fault case DB6, learns an initial model of the fault calculation estimation model DB8, and generates a learned fault calculation estimation model DB8. the fault state change estimation step in which the fault state change estimation unit 4 of the power system stabilization system 1 receives as input the system state, the system model DB5, the learned fault calculation estimation model DB8, and the fault case DB6, and calculates a fault state change estimation value DB10; the fault state estimation step in which the fault state estimation unit 5 of the power system stabilization system 1 estimates a fault state based on the fault state change estimation value DB10 and the system state DB9; and a display step in which the display unit 93 of the power system stabilization system 1 displays one or more of the input / output data.

[0084] In this way, the fault state change amount DB7 calculated by the fault state change amount calculation unit 2 is input as a target variable to the trained fault calculation estimation model DB8 having an output layer DB81 that limits the output range, so that the range of the fault state change amount estimated value DB10 estimated by the fault state change amount estimation unit 4 is limited, thereby making it possible to reduce the maximum prediction error by artificial intelligence and speed up fault calculation.

[0085] In this embodiment, in addition to the system model DB5, the pre-fault state DB3, and the fault case DB6, the fault calculation estimation model learning unit 3 further inputs the learned fault calculation estimation model DB8 as an explanatory variable and inputs the fault state change amount DB7 as a predictive variable to learn the fault calculation estimation model DB8. In this way, it is possible to generate a fault calculation estimation model DB8 with better prediction performance.

[0086] In this embodiment, the fault state change amount DB7 includes one or more of the amount of change in voltage, the amount of change in phase, the amount of change in phase difference, the amount of change in active power, and the amount of change in reactive power. In this way, the fault calculation estimation model learning unit 3 can learn the fault calculation estimation model DB8 while taking into consideration one or more of the amount of change in voltage, the amount of change in phase, the amount of change in phase difference, the amount of change in active power, and the amount of change in reactive power.

[0087] In this embodiment, the fault calculation and estimation model DB8 includes an output layer that limits the output range, so that the range of prediction errors can be narrowed by the output layer.

[0088] In this embodiment, the fault calculation estimation model learning unit 3 learns in a direction to minimize one or more of the maximum error between the predicted value of the fault state change amount for calculating the fault calculation estimation model DB8 and the fault state change amount, the square sum error between the predicted value of the fault state change amount for calculating the fault calculation estimation model DB8 and the fault state change amount, and the sum of the absolute error value between the predicted value of the fault state change amount for calculating the fault calculation estimation model DB8 and the fault state change amount. In this way, it is possible to create a fault calculation estimation model DB8 with higher prediction accuracy.

[0089] In this embodiment, the fault calculation estimation model learning unit 3 judges the validity of the learning result for the learned fault calculation estimation model, and uses at least one of the following as a criterion for judging the validity of the learning result: setting a threshold value for the evaluation function value; and evaluating the total calculation time that can be reduced by performing a simulated online calculation using the learning result. In this way, the online calculation time can be reduced.

[0090] In this embodiment, when learning of the learned fault calculation estimation model DB8 is insufficient, the fault calculation estimation model learning unit 3 uses one or more of the following means: changing the structure of the fault calculation estimation model DB8, changing the learning parameters of the fault calculation estimation model DB8, increasing the input data of the fault calculation estimation model DB8, decreasing the input data of the fault calculation estimation model DB8, and changing the format (expression method) of the input data of the fault calculation estimation model DB8. In this way, a fault calculation estimation model DB8 with higher prediction accuracy can be generated.

[0091] In this embodiment, the fault calculation countermeasure unit 7 performs calculations to plan countermeasures for the fault calculation results by inputting the post-fault state DB 11 online, and the control unit 8 is an example of a fault calculation unit, and executes fault countermeasures according to the contents of the fault calculation results. In this way, it is possible to execute appropriate fault countermeasures according to the countermeasures that have been planned.

[0092] In this embodiment, the correction unit 6 generates an online corrected post-fault state DB11 based on the system model DB5 and the fault state estimated by the fault state estimation unit 5. In this way, the fault calculation countermeasure unit 7 can plan a fault countermeasure using a more accurate post-fault state DB11.

[0093] Although the present embodiment has been described above, the technical scope of the present invention is not limited to the scope described in the above embodiment. It is clear from the claims that various modifications and improvements to the above embodiment are also included in the technical scope of the present invention. [Industrial Applicability]

[0094] The present invention can be applied to a power system stabilization system that uses artificial intelligence to stabilize a power system. [Explanation of symbols]

[0095] GR1...power system, G1~G3...generators, LD1~3...loads, B1~8...busbars, L1~L9...transmission lines, 1...power system stabilization system, 2...fault state change amount calculation unit, 3...fault calculation estimation model learning unit, 4...fault state change amount estimation unit, 5...fault state estimation unit, 6...correction unit, 7...fault calculation countermeasures unit, 8...control unit, 91...input unit, 92...calculation unit, 93...display unit, 94...memory, 95...communication unit, 96...communication network, 97...measuring instruments, 98...control terminal.

Claims

1. a failure state change amount calculation unit that calculates a failure state change amount using a pre-failure state and a post-failure state as input; a fault calculation estimation model learning unit that receives the fault state change amount, a system model, the pre-fault state, and a fault case, learns an initial model of a fault calculation estimation model, and generates a learned fault calculation estimation model; a fault state change amount estimating unit that receives as input a system state, the system model, the fault calculation / estimation model after learning, and the fault case, and calculates a fault state change amount estimate value; a fault state estimation unit that estimates a fault state based on the fault state change amount estimate value and the system state; a display unit for displaying one or more of the input and output data; A power system stabilization system comprising:

2. The fault calculation and estimation model learning unit includes: In addition to the system model, the pre-fault state, and the fault case, the learned fault calculation estimation model is input as an explanatory variable, and the fault state change amount is input as a predictive variable, thereby learning the fault calculation estimation model.

2. The power system stabilization system according to claim 1 .

3. The fault state change amount is The change in voltage, the change in phase, the change in phase difference, the change in active power, and the change in reactive power are included in the above.

3. The power system stabilization system according to claim 2 .

4. The fault calculation estimation model includes an output layer that limits an output range.

4. The power system stabilization system according to claim 3 .

5. The fault calculation and estimation model learning unit includes: learning in a direction to minimize at least one of a maximum error between the predicted value of the fault state change amount for calculating the fault calculation estimation model and the fault state change amount, a square sum error between the predicted value of the fault state change amount for calculating the fault calculation estimation model and the fault state change amount, and a sum of absolute values ​​of errors between the predicted value of the fault state change amount for calculating the fault calculation estimation model and the fault state change amount; 5. The power system stabilization system according to claim 4 .

6. The fault calculation estimation model learning unit judges the usefulness of a learning result related to the learned fault calculation estimation model, and uses, as a criterion for judging the usefulness of the learning result, at least one of setting a threshold value for an evaluation function value and evaluating a total calculation time that has been reduced by performing a simulated online calculation using the learning result.

6. The power system stabilization system according to claim 5 .

7. When learning of the already-learned fault calculation estimation model is insufficient, the fault calculation estimation model learning unit performs one or more of the following: changing a structure of the fault calculation estimation model; changing a learning parameter of the fault calculation estimation model; increasing input data of the fault calculation estimation model; decreasing input data of the fault calculation estimation model; and changing a format of the input data of the fault calculation estimation model.

7. The power system stabilization system according to claim 6 .

8. a fault calculation countermeasure unit that performs an online calculation using the post-fault state as an input to plan a countermeasure against the fault calculation result; a fault calculation unit that executes fault calculation according to the fault calculation result; The power system stabilization system according to claim 7, further comprising:

9. a correction unit that calculates an online corrected post-fault state based on the system model and the fault state estimated by the fault state estimation unit, 2. The power system stabilization system according to claim 1 .

10. A fault state change amount calculation step in which a fault state change amount calculation unit of the power system stabilization system calculates a fault state change amount by inputting a pre-fault state and a post-fault state; a fault calculation estimation model learning step in which a fault state change amount, a system model, the pre-fault state, and a fault case are inputted to a fault calculation estimation model learning unit of the power system stabilization system, an initial model of the fault calculation estimation model is learned, and a fault calculation estimation model after learning is generated; a fault state change amount estimation step in which a fault state change amount estimation unit of the power system stabilization system receives a system state, the system model, the fault calculation and estimation model after learning, and the fault case as inputs, and calculates a fault state change amount estimation value; a fault state estimation step in which a fault state estimation unit of the power system stabilization system estimates a fault state based on the fault state change amount estimate value and the system state; A display step in which a display unit of the power system stabilization system displays one or more of the input / output data; A power system stabilization method comprising:

Images