JP2024006734A - Power system stabilization device and method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an index capable of performing the same evaluation on stabilization effects due to electrical control of a synchronous generator and a renewable energy power source, and an electrical control algorithm using the same.

SOLUTION: A power system stabilization device determines an electrical control target in the case of fault occurrence in accordance with stability in the case of assumed fault in a power system including a synchronous generator and a renewable energy power source. The power system stabilization device comprises: an electrical control effect estimation model creation section for creating an electrical control estimation model through recursive and machine learning based on a transient stability calculation result when performing electrical control in the case of estimated fault in the power system; an electrical control effect estimation calculation section for estimating an electrical control effect due to electrical control on one or more of the synchronous generator and the renewable energy power source based on the electrical control effect estimation model and the transient stability calculation result; an electrical control target selection section for selecting, for electrical control, one of the synchronous generator and the renewable energy power source based on an estimation calculation result of the electrical control effect and selecting it as the electrical control target until transient stability of the power system is stabilized; and an output section for outputting results of the respective sections.

SELECTED DRAWING: Figure 1

COPYRIGHT: (C)2024,JPO&INPIT

Description

The present invention relates to a power system stabilizing device and method for maintaining power system stability.

When a failure (accident) due to a lightning strike or the like occurs in an electric power system, the output of some synchronous generators (SGs) may become unstable. If effective measures are not taken against this unstable phenomenon, the number of unstable synchronous generators will increase over time, and there is a risk that it will eventually lead to a major power outage.

To address the above-mentioned problem, conventional power system stabilization devices have maintained the transient stability of the power system by cutting off unstable synchronous generators from the power system in the event of a failure, in other words limiting the power supply (electrification). . For example, in the power system stabilization device described in Non-Patent Document 1, in order to maintain transient stability in the event of a hypothetical failure (hypothetical accident) in the electric power system, a synchronous generator to be shut down is calculated in advance by numerical simulation, and a hypothetical failure is detected. When this actually occurs, stabilization control is performed to shut down the synchronous generator. This power system stabilization device attempts to maintain transient stability by cutting off the synchronous generator whose internal phase difference angle is large in the event of a failure. In addition, stabilization control includes not only the above-mentioned power restriction but also a method of suppressing or stopping the output of the generator, but hereinafter, the term "power control" includes the output restriction and stopping of the generator.

On the other hand, in recent years, in order to reduce CO2 emissions, the introduction of renewable energy sources (RES) such as solar power generation and wind power generation has been actively promoted as an alternative to synchronous generators such as thermal power generation. , the proportion of synchronous generators in the power system is on the decline. However, if the proportion of synchronous generators decreases, the inertia and voltage maintenance ability of the power system that they have traditionally served will decrease, leading to a risk of a decrease in transient stability.

Further, the reduction in the number of synchronous generators means the reduction in the number of stabilizing means in the conventional power system stabilization device. Renewable energy power sources do not have an internal phase difference angle, which is an indicator for selecting a shedding machine in conventional power system stabilization devices, so conventional power system stabilization devices cannot select renewable energy power sources as targets for shedding. . Therefore, it may be difficult to maintain transient stability using conventional power system stabilization devices, and the total amount of power shedding required to maintain transient stability may increase (increase in cost required to recover from power grid). be.

To address the above-mentioned problems, adding renewable energy power sources, which are expected to be introduced more frequently in the future, as candidates for power shedding increases the possibility of maintaining transient stability and reduces the total amount of power shedding required for stabilization ( This can be expected to reduce the costs required for returning from electricity grids.

In this regard, Patent Document 1 is known as an invention related to a power system stabilizing device that includes renewable energy power sources as power grid candidates. Patent Document 1 proposes to provide a power system stabilization device that can select a renewable energy power generation device (renewable energy power source) whose output amount is reduced or stopped, and efficiently ensure the stability of the power system. A stability calculation unit 11 that calculates stability regarding assumed target accident cases, an acceleration tendency synchronous generator extraction unit 12 that extracts one or more acceleration tendency synchronous generators 4, and a renewable energy A sensitivity calculation unit 13 that calculates the deceleration sensitivity of the synchronous generator 6 with an acceleration tendency when the output amount of the power generation device 6 is reduced by the first reduction amount P1, and the stability of the extracted renewable energy power generation device 6. An electric power system comprising: a renewable energy reduction amount addition unit 14 that determines the total reduction amount P3 of the output amount that is determined to be stable; and a reduction target determination unit 15 that determines the renewable energy power generation device 6 whose output amount is to be reduced. Stabilizer."

In addition, Patent Document 2 aims to provide a power system stabilization system that can efficiently ensure the stability of the power system by selecting a renewable energy power generation device that reduces or stops the output amount. A stabilization index change amount calculation unit that calculates the amount of change in the stabilization index of synchronization stability when reduction occurs, and multiple renewable energy power generation devices are classified into groups based on the calculated amount of change in the stabilization index. and an estimation model creation section that calculates the amount of decrease in output power in the event of an assumed accident for each group for each power flow cross section, and constructs an estimation model that estimates the amount of decrease for each group. and a control amount distribution unit that allocates the reduction amount for each group to the renewable energy power generation devices in the group. A power system stabilization system comprising a control unit 20 that instructs a renewable energy power generation device to reduce the amount of power.

JP2020-96472A JP2021-69214A

Hiroshi Yoshida, Ryuji Tate, Koya Takafuji, Hironori Imaeda, Masaru Takeishi, Hiroyuki Taguchi, Kenichiro Kusaba: "Development of an integrated online grid stabilization system (ISC) adapted to the next generation grid", Electrical Engineering Theory B, Vol. ．． 137, No. 6, pp. 434-445 (2017)

Patent Document 1 discloses that in a power system with a large number of renewable energy power sources connected and a small number of synchronous generators in operation, if the synchronous generators are shut down in the event of a failure, the power system may become unstable. Based on the background that there is This method uses a method to select the target of the synchronous generator's power restriction when power restriction alone is not sufficient.

However, depending on the region where the failure occurs, the proportion of renewable energy power sources in the power system may be small, or there may be few renewable energy power sources that have a high stabilizing effect through power outages. The method disclosed in Patent Document 1, which selects the target of power cut-off of renewable energy power sources with priority over other machines, increases the total amount of power cut required for stabilization (increases the cost required to recover from power cut-off), and reduces excessive power consumption. There is a possibility that problems such as a negative impact on the power grid due to the curtailment and an increase in the amount of calculation required to select the curtailment machine may occur.

Therefore, there is a need for a method that can prioritize the selection of synchronous generators or renewable energy power sources to be cut off based on those with the highest stabilizing effect, regardless of the proportion of renewable energy sources in the power system.

In view of the above, an object of the present invention is to provide an index that can equally evaluate the stabilizing effect of power shedding of a synchronous generator and a renewable energy power source, and a power shedding algorithm using the index.

Based on the above, in the present invention, "the power control target that performs stabilization control in the event of a failure in the power system is determined according to the stability at the time of an assumed failure in the power system including synchronous generators and renewable energy power sources. A power system stabilization device that is defined, and creates a power shedding effect estimation model using a regression formula and machine learning based on the transient stability calculation results when performing power shedding in the event of a hypothetical failure in the power system. a shedding effect estimation calculation unit that estimates the shedding effect due to shedding of one or more of the synchronous generator and the renewable energy power source based on the shedding effect estimation model and the transient stability calculation results; Based on the magnitude of the calculation results, one of the synchronous generators and renewable energy power sources that are candidates for electricity reduction is selected, and electricity reduction is performed based on the calculation results for estimating the effect of electricity reduction until the transient stability of the power system is stabilized. An electric power system stabilizing device characterized by comprising: a power restriction target selection unit that selects a candidate as a power restriction target; and an output unit that outputs the results of each unit.

In addition, in the present invention, "a power system that determines the target of power outage for implementing stabilization control in the event of a failure in the power system according to the stability at the time of an assumed failure in the power system including a synchronous generator and a renewable energy power source" The stabilization method includes a step of creating a power shedding effect estimation model using a regression formula and machine learning based on the transient stability calculation results when performing power shedding in the event of an assumed failure in the power system, and a power shedding effect estimation model creation step. A shedding effect estimation calculation step of estimating the shedding effect due to one or more shedding of the synchronous generator and the renewable energy power source based on the shedding effect estimation model and the transient stability calculation results; Select one of the synchronous generators and renewable energy power sources that are candidates for power shedding based on the size, and select the candidates for power shedding based on the calculation results for estimating the power shedding effect until the transient stability of the power system is stabilized. A method for stabilizing a power system, comprising: a step of selecting an electrical power reduction target to be controlled; and an output step of outputting the results of each part.

According to the present invention, the stabilizing effect of power shedding of a synchronous generator and a renewable energy power source is evaluated in the same way, and at least one kind of the synchronous generator or renewable energy power source is sheared in order of the stabilizing effect. Since this option can be selected, it is possible to improve transient stability regardless of the proportion of renewable energy power sources in the power system, and to reduce the total amount of power outage required for stabilization (the cost required to recover from power outage). Moreover, the amount of calculations for the stabilizing device can be reduced.

1 is a diagram showing an example of a functional configuration of a power system stabilizing device 10 according to a first embodiment of the present invention. 1 is a diagram showing an example of a hardware configuration of a power system stabilizing device 10 according to a first embodiment of the present invention. The figure which showed the program which program database DB1 holds. The figure which shows an example of the information D13a regarding system configuration among system equipment data D13. The figure which shows an example of the information D13b regarding system model data among system equipment data D13. The figure which shows an example of system measurement data D12 which system measurement database DB14 holds. The figure which shows an example of the assumed failure data D15 which the assumed failure database DB15 holds. A diagram showing an example of electricity restriction data D16 held by electricity restriction database DB16. The figure which shows an example of the threshold value data D17 which threshold value database DB17 holds. The figure which shows an example of the threshold value data D17b used for evaluation of the electricity restriction effect estimation model among the threshold value data D17 which the threshold value database DB17 holds. FIG. 3 is a diagram showing the overall flow of processing of the power system stabilization device 10 according to the first embodiment of the present invention. 12 is a diagram showing a detailed flow of the electricity restriction target selection process in processing steps S33 to S35 of FIG. 11. FIG. FIG. 12 is a diagram showing a detailed flow of the electricity cut effect estimation model creation process in processing step S31 of FIG. 11; The figure explaining the image of deceleration energy estimation by machine learning. A diagram for explaining the application of sparse regression, etc. to improve accuracy. The figure which shows the image when the objective variable is the deceleration energy deviation before and after electrical control. FIG. 6 is a diagram in which the objective variable is the maximum value δSGjmax of the internal phase difference angle after electrical control. FIG. 6 is a diagram in which the target variable is the maximum value deviation ΔδSGjmax of the internal phase difference angle after electrical control. The figure which shows the image of the hyperparameter (alpha) which is one of the learning setting value data D12 which the learning setting value database D12 holds. FIG. 3 is a diagram showing an image of changes in the regression coefficients of each variable with respect to changes in the hyperparameter α during search. The figure which shows the detailed flow which newly calculates the transient stability calculation result data D11 at the time of electrical restriction. FIG. 3 is a diagram showing an example of the creation result of an electricity restriction effect estimation model. The figure which shows the time change of the active power Pj of the electrical power reduction candidate SGj from before a fault occurs until the fault occurs and the electricity is cut off. A diagram showing an example in which a sparse regression (reduced regression, sparse) model is also applied. Comparing the changes in the active power P _SGj of the synchronous generator SGj and the internal phase difference angle δSGj of the synchronous generator SGj from before the failure occurs until the failure occurs and the power is cut off when a sparse regression model is also applied. Figure shown. The figure which shows an example of a power cut effect estimation result. The figure which shows an example of error distribution evaluation at the time of learning of an electricity control estimation model. A diagram showing details of learning and an image of estimation calculation. FIG. 3 is a diagram showing an example of result display by the display unit 11. FIG. FIG. 3 is a diagram showing an example of result display by the display unit 11. FIG. The figure which shows the example of a functional structure of the power system stabilization device 10 based on Example 2 of this invention. The figure which shows the hardware configuration example of the power system stabilization device 10 based on Example 2 of this invention. The figure which shows the processing content of the electricity restriction target selection part 34d.

Embodiments of the present invention will be described below with reference to the drawings. In the description of the present invention, "stability" is used synonymously with "stability", and in the following description, it is uniformly expressed as "stability".

First, a functional configuration example of a power system stabilizing device 10 according to a first embodiment of the present invention will be described using FIG. 1. In addition, in the following explanation, when the operating body is written as "○○ part", the processor 14 in FIG. The functions of the department shall be realized.

The power system stabilization device 10 includes input data D1 that is stored in advance by a power system operator or planner, a calculation unit 41 that determines power system stabilization control based on the input data D1, and a calculation unit 41 that determines power system stabilization control based on the input data D1. It is composed of output data D2 which is a calculation result, a display section 11 which has an output function to display the contents of input data D1 and output data D2, and a communication section 13.

The control device 3 (slave station) receives the electricity restriction target calculation result table data D25 from the power system stabilization device 10 via the communication network 200. In addition, when a failure occurs in the power system 100, the control device 3 compares the failure information acquired from the measurement devices 30a to 30b with the power cutoff target calculation result table data D25, and controls the generator terminal station to shut off the generators 110a to 110c. (transfer cutoff device).

The database DB that stores the input data D1 includes a transient stability calculation result database DB11 during electric power system that holds transient stability calculation result data D11 during electric power system, a learning setting value database DB12 that holds learning setting value data D12, and system equipment. System equipment database DB13 (including system model) that holds data D13, system measurement database DB14 that holds system measurement data D14, assumed failure database DB15 that holds assumed failure data D15, and electric power system that holds electrical system restriction data D16. It is composed of a constraint database DB16 and a threshold database DB17 that holds threshold data D17.

The calculation unit 41 includes a calculation unit 41a and a calculation unit 41b. Since these have different calculation loads, they are preferably configured as separate servers. For example, the calculation unit 41a is preferably executed offline, or the calculation unit 41b is preferably executed online at a different cycle. In the following description, it is assumed that the processing of the arithmetic unit 41a is pre-executed off-line, and that the processing of the arithmetic unit 41b is executed online based on this result. Note that the calculation unit 41a executes the processing of the electricity shedding effect estimation model creation unit 31, and the calculation unit 41b executes the processing of the transient stability calculation unit 32, electricity shedding effect estimation calculation unit 33, electricity shedding target selection unit 34, and grid stabilization confirmation. The processing of the unit 35 is executed in this order.

The database DB that stores the output data D2 is the result of the power shedding effect estimation model database DB21 that stores the power shedding effect estimation model data D21 that is the result of the power shedding effect estimation model creation section 31, and the results of the transient stability calculation section 32. A transient stability calculation result database DB22 that stores transient stability calculation result data D22, a power shedding effect estimation calculation result database DB23 that stores power shedding effect estimation calculation result data D23 that is the result of the shedding effect estimation calculation section 33, A grid stabilization confirmation result database DB24 that stores grid stabilization confirmation result data D24, which is the result of the grid stabilization confirmation unit 34, and a grid stabilization target that saves the grid stabilization confirmation result data D25, which is the result of the grid stabilization confirmation unit 35. It consists of a selection result database DB25.

FIG. 2 is a diagram showing the hardware configuration of the power system stabilizing device 10 of the first embodiment. It is a figure showing an example. The upper part of FIG. 2 shows an example of the hardware configuration of the power system stabilizing device 1, and the lower part shows an example of the configuration of the power system that is the object of control.

The power system stabilizing device 10 includes various input databases DB11 to DB17, a program database DB1, various output databases DB21 to DB25, a display section 11, an input section 12, a communication section 13, a processor 14, a memory 15, and the like. It is constituted by a connecting bus line 43.

FIG. 3 shows a program group stored in the program database DB1. The program database DB1 includes a power shedding effect estimation model creation program P31, a transient stability calculation program P32, a power shedding effect estimation calculation program P33, a power shedding target selection program P34, and a grid stabilization confirmation program P35.

The display unit 11, which has an output function, is composed of, for example, one or more of a display device, a printer device, a projector device, an audio output device, and the like. The display unit 11 displays one or more of various input data D11 to D17 and various output data D21 to D25 on a screen. An example of the screen to be displayed will be described later.

The input unit 12 includes, for example, one or more of a keyboard, a switch, a mouse, a touch panel, a voice input device, and the like. The input unit 22 inputs various conditions for operating the power system stabilizing device 10.

The communication unit 13 exchanges data with the power system 100 and the control device 3 via the communication network 200. The communication unit 13 includes a circuit and a communication protocol for connecting to the communication network 200. The communication network 200 may be a leased line, a WAN (Wide Area Network) such as the Internet, or a LAN (Local Area Network) such as WiFi or Ethernet (registered trademark). However, WAN and LAN may coexist.

The processor 14 reads programs necessary for the processing of the calculation unit 41 from among the various programs that make up the program database DB1, and executes calculations. It also performs processing related to searching data in various databases stored in the storage device 26 and instructing display of processing results. The processor 14 may be a CPU (Central Processing Unit) or a GPU (Graphics Processing Unit). Processor 14 may be a single-core processor or a multi-core processor. The processor 24 may include a hardware circuit (eg, FPGA (Field-Programmable Gate Array) or ASIC (Application Specific Integrated Circuit)) that performs part or all of the processing. Processor 14 may include a neural network. Processor 14 may be configured as one or more semiconductor chips or may be configured as a computing device such as a calculation server.

The memory 15 is composed of a storage device such as a RAM (Random Access Memory), and stores programs read from the program database DB1, input data D1, output data D2, etc., and also provides a work area necessary for each process. The data may also be provided to the processor 14.

Each database DB is a storage device with a large storage capacity, such as a hard disk drive or SSD (Solid State Drive). The storage device 26 can hold execution files of various programs and data used to execute the programs. Each program may be software that can be installed in the power system stabilization device 10, or may be incorporated in the power system stabilization device 10 as firmware.

The power system 100 as an example shown in the lower part of FIG. 2 includes generators 110a to 110c such as synchronous generators and renewable energy power sources, nodes (busbars) 120a to 120c and 121a to 121c, transformers 130a to 130c, and branches. (Line) Consists of 140a to 140c, etc. The generators mentioned here include, for example, synchronous generators such as thermal power generators, hydroelectric power generators, or nuclear power generators, and renewable energy power sources such as solar power generators, solar thermal power generators, wind power generators, and wind farms. , tidal current power generation, and other power sources that are connected to the grid via an inverter, and those that are connected to the grid via HVDC are also considered renewable energy power sources.

Furthermore, the power system 100 includes measuring devices 30a to 30b. The measuring devices 30a to 30b cooperate with the power system stabilization device 10 via the communication network 200.

The measuring devices 30a to 30b measure the output P of the generators 110a to 110c, the voltage value V at each node 120a to 120c and 121a to 121c, and the active power P and reactive power Q flowing through the transformers 130a to 130c and branches 140a to 140c. Acquire one or more of TM (Telemeter) information such as values, SV (Super Vision) information such as on/off information of circuit breakers such as nodes, transformers, branches, and phase adjustment equipment as system measurement data D14, and send it to the communication section 13. Send. Specifically, the measurement is performed using a voltage transformer (VT), a potential transformer (PT), a current transformer (CT), or the like.

The power system stabilization device 10 can periodically acquire TM information and SV information from the measurement devices 30a to 30b via the communication network 300 and store them in the system measurement value DB.

Further, some of the measuring devices 30a to 30b detect information such as the location and condition of the fault when a fault occurs in the power system 100, and transmit the information to the control device 3 via the communication network 200.

The control device 3 (slave station) receives the electricity restriction target calculation result table data D25 from the power system stabilization device 10 via the communication network 200. In addition, when a failure occurs in the power system 100, the control device 3 compares the failure information acquired from the measurement devices 30a to 30b with the power cutoff target calculation result table data D25, and controls the generator terminal station to shut off the generators 110a to 110c. (transfer cutoff device).

Here, the data D11 to D17 stored in the various input databases DB11 to DB17 constituting the power system stabilizing device 10 will be explained using texts and drawings, respectively.

The electrical braking transient stability calculation result database DB11 stores electrical braking results (transient stability calculation result data D11 during electrical braking) in transient stability calculations that have been performed in advance and will be described later. Note that the transient stability calculation result data D11 during power reduction may be obtained by preprocessing or may be generated by creating a learning model off-line.

The learning setting value database DB12 stores parameters and setting values (learning setting value data D12) necessary for machine learning in creating the electricity restriction effect estimation model. For example, the target of explanatory variables of the regression model described later, the target of objective variables, the hyperparameter α required for sparse regression analysis, the number of divisions for cross validation, etc.

One of the system equipment data D13 stored in the system equipment database DB13 is the system configuration of the power system (the connection relationship of one or more of the system bus, line, power source, load, transformer, and each control device) and the line. In addition to information regarding impedance (R + j

Another type of system equipment data D13 stored in the system equipment database DB13 is system model data D13b, which includes synchronous generators and renewable energy power sources that are necessary for numerical analysis of power systems using computers. , information regarding the load model, etc. The system model data D13b includes specifications such as slack nodes and convergence determination thresholds for performing transient stability calculation processing.

FIG. 4 shows an example of information D13a regarding the system configuration among the system equipment data D13 held by the system equipment database DB13. Regarding the system configuration information D13a, the system equipment database DB13 stores information on loads, renewable energy power sources, transformers, phase adjustment equipment, etc. in addition to information on branches (lines) and synchronous generators that make up the power system. ing. Note that the data D13a to be held for these devices includes the transmission line number, number of lines, node numbers at both ends, resistance, reactance, etc. in the case of a branch, and the synchronous generator number, linked node, and the like in the case of the same device. It is good to have the number of parallel units, rated capacity, rated output, reactance, etc., and also in the case of loads, renewable energy power sources, transformers, phase adjustment equipment, etc., appropriate information should be held as appropriate.

FIG. 5 shows an example of information D13b related to system model data among the system equipment data D13 held by the system equipment database DB13. Regarding information D13b, the system model database DB13 includes information on model types and constants used here regarding synchronous generators, renewable energy power sources, and load models that are necessary for numerical analysis of power systems using computers and calculators. is stored.

FIG. 6 shows an example of the system measurement data D14 held by the system measurement database DB14. The system measurement database DB14 includes the outputs of the generators 110a to 110c of the power system 100, voltage values at each node 120a to 120c and 121a to 121c, transformers 130a to 130c, and branches 140a to 121c, acquired via the communication network 200. One or more of the values of active power and reactive power flowing through 140c, on/off information of circuit breakers such as nodes, transformers, branches, and phase adjustment equipment are stored. Active power P, reactive power Q, voltage V, phase angle δ, current I, power factor Φ, tap value, on/off information of switches between the power system and nodes, branches, transformers, SCs, ShRs, etc. at various points in the system etc. are included. This information is stored in chronological order in association with time stamp information of measurement date and time for each measurement location.

In addition, the system equipment data D13, system measurement data D14 (data with a time stamp or PMU data may be used), and various setting values may be obtained from a supervisory control device, central power dispatch center, or EMS, or may be input manually. You can. When manually inputting information, the information is manually input using the input unit 12 and stored. Note that during input, necessary image data is generated by the processor 14 and displayed on the display unit 11. When inputting data, you can use the completion function and do it semi-manually so that you can set a large amount of data.

FIG. 7 shows an example of assumed failure data D15 held by the assumed failure database DB15. The hypothetical failure database DB15 stores information regarding the location, aspect, failure removal means, etc. of hypothetical failures in the power system.

FIG. 8 shows an example of electricity restriction data D16 stored in electricity restriction database DB16. The power restriction restriction database DB16 stores information regarding the synchronous generator and the renewable energy power source, such as whether or not the power can be restricted to the generator, and if the power restriction is possible, the timing thereof.

FIG. 9 shows an example of threshold data D17a stored in the threshold database DB17. In the illustrated example, an internal phase difference angle threshold is stored in the threshold database DB17 as threshold data D17a used for acceleration tendency determination and stability determination.

Further, the threshold value database DB17 holds threshold value data D17b used for evaluating the power reduction effect estimation model created in the calculation unit 41a, an example of which is shown in FIG. In order to evaluate the shedding effect estimation model, the coefficient of determination R ² , the maximum and minimum values of the error deviation of the test results (deceleration energy estimate) using the learning model, and the maximum and minimum absolute values of the error deviation are used. A threshold value necessary to determine whether the value, root mean square error RMSE, etc. is large or small is used, and a specific numerical value such as 0.900, for example, is stored as the threshold value in the threshold value database DB17.

Here, we will explain the coefficient of determination ^R2 for evaluating the degree of adaptation when the electricity restriction effect estimation model is obtained as a regression model. This can be expressed by equation (1). However, in equation (1), Δmi is the estimation error, N is the number of verification cross sections, and Var(m) is the variance.

According to equation (1), the closer the coefficient of determination ^R2 is to 1, the higher the accuracy of the regression equation is (it becomes 1 when the mean square error is 0), and it is closer to 1 for both training data and test data. It can be said that the model is more versatile. There is no statistical standard that says that the accuracy is high if the coefficient of determination ^R2 is higher than that, but for example, if the coefficient of determination is 0.5 to 0.8, the accuracy of the multiple regression equation is slightly high, 0. If it is between .8 and 1.0, it is judged to be very high. The threshold value data D17b used for model evaluation is determined using this appropriate numerical value as a threshold value.

Similarly, the maximum and minimum values of the error deviation between the estimated value and the true value (verification data) of the regression model, the maximum and minimum values of the absolute value of the error deviation, and the root mean squared error RMSE (Root Mean Squared Error ) Alternatively, an appropriate value for the mean absolute error MAE (Mean Absolute Error) is also determined as a threshold value and used for evaluation. Here, RMSE (root mean square error) can be evaluated using equation (2).

Next, the processing contents of the calculation unit 41 will be explained using FIG. 11. FIG. 11 shows the overall flow of processing of the power system stabilizing device 10 in the first embodiment. The flow of arithmetic processing will be explained for each processing step.

In processing step S31, which is a process of the calculation unit 41a in FIG. 1 and is preferably executed off-line, an electric shearing effect estimation model is The creation program P31 is executed to create the electricity cutoff effect estimation model data D21 of the power system 100.

The following is the processing of the calculation unit 41b in FIG. 1. First, in processing step S32, any generator is The transient stability calculation program P32 is executed under the condition of no electricity grid (no electricity grid), and the transient stability calculation result data D22 of the power system 100 is output.

Specifically, in processing step S32a, a hypothetical failure is set, and in processing step S32b, an uncontrolled transient stability calculation (including various data reception and state estimation processing) for the set failure is performed, In processing step S32c, it is determined whether the power system 100 is transiently stable or unstable based on the transient stability calculation result data D22 and the threshold data D17. The stability is determined using, for example, the internal phase difference angle, frequency, voltage, etc. of the synchronous generator as indicators. If stable, the process returns to step S32a.

Note that if all failures have been set, this processing flow ends. In the case of instability, the process moves to processing step S33, which will be described later.

Here, the flow up to the transient stability calculation will be explained in detail. First, the system measurement data D14, system equipment data D13, and setting value data are read and used to create system power flow cross-sectional data. System power flow cross-sectional data can also be created by finding missing information through state estimation and power flow calculations.

Here, we will provide some supplementary information about state estimation and power flow calculation. Using the system measurement data D14, system equipment data D13, and calculation setting data, the system state at the time of system measurement is calculated by a state estimation calculation/power flow calculation program not shown in the drawing, and is stored in the system measurement database 14. do.

Note that state estimation calculation is the process of determining and removing abnormal data from observation data based on observation data of power transmission and distribution equipment such as substations, power plants, and power transmission lines, as well as connection data. It is a computational function that estimates a plausible system state over a specific time section. Here, the state estimation calculation is described, for example, by Lars Holten, Anders Gjelsvlk, Sverre Adam, F. F. Wu, and Wen-Hs Iung E. Liu, Comparison of Different Methods for State Estimation, IEEE Transaction on Power Systems, Vol. 3 (1988), pp. It can be carried out according to various methods described in 1798-1806.

In addition, in the power flow calculation, the voltage V of each node 120 and the output command values P and Q of the load necessary for the power flow calculation of the state estimation result, the system equipment data D13, and the calculation setting data are used to A node, a synchronous phase modifier, and a reactive power compensator are designated as PV, a substation node and a load node are designated as PQ, and a preset node voltage V and phase angle θ are designated for a preset slack node in the power system 100. It is preferable to perform power flow calculation using the Newton-Raphson method together with the admittance matrix Yij created from the system equipment data 13, and to store the calculation results.

Here, the power flow calculation method is, for example, WILLIAM F TINNEY, CLIFFORD E HART, Power Flow Solution by Newton's Method, IEEE Transaction on Power APPARATUS AND S YSTEMS, VOL. PAS-86, NO. 11 (1967) pp. This can be carried out in accordance with the method of No. 1449-1967.

The power flow calculation method is based on the AC method. In addition, based on the power flow cross section data, assumed failure data D15, and system equipment data D13 created through state estimation and power flow calculations, transient stability calculations were carried out by Prabha Kundur, Power System Stability and Control, The Epri Power System Engineering ( 1994 )pp. 827-954, development of a comprehensive stability analysis system for large-scale power systems, Central Research Institute of Electric Power Industry General Report T14 (1990), analysis and operation technology that supports the use of power systems, IEEJ Technical Report No. 1100 (2007) )pp. This can be carried out in accordance with various methods shown in 106-110 and the like.

Next, in processing step S33, the electrical shedding effect estimation calculation program P33 is executed using the transient stability calculation result data D22 and the electrical shedding effect estimation model data D21, and the electrical shedding effect estimation calculation result data D23 is output.

Next, in processing step S34, the electricity restriction target selection program P34 (which includes stabilization effect confirmation processing by stability calculation during electricity restriction) uses the electricity restriction data D16 and the electricity restriction effect estimation calculation result data D23. ), and then, in processing step S35a, the transient stability calculation program P32 is executed in a state in which electrical outage is selected, and the transient stability calculation result during electrical outage of the power system 100 is output.

In processing step S35b, a grid stabilization confirmation program P35 is executed on this result and the threshold value data D17, and grid stabilization confirmation result data D24 and a power cut target selection result table D25 are output.

Here, if it is determined that the shedding is unstable, the process returns to step S33, and in order to obtain the next shedding machine, the shedding effect estimation calculation is performed again, the next shedding machine is selected, and the transient stability is stabilized. The process step S35b is reached by calculating the degree.

Further, if it is determined that the process is stable, it is determined in processing step S35c whether the selection of all possible failures has been completed, and if the selection has not been completed, the process returns to processing step S32a. If the process has ended, the process advances to step S36.

In processing step S36, the electricity restriction target selection result table data D25 obtained up to processing step S35c is transmitted to the control device 3 using the transmitter 36. For example, in consideration of the amount of calculation by the calculation unit 2, this transmission is performed periodically at a period of 30 seconds. Further, in processing step S37, each output data is displayed on the screen.

FIG. 12 shows a detailed flow of the electricity restriction target selection process in processing steps S33 to S35 in FIG. 11. The procedure for selecting an electrical restriction target will be explained for each processing step in FIG. 12.

In processing step S33a, the electrical shedding effect estimation calculation program P33 is executed using the transient stability calculation result data D22 and the electrical shedding effect estimation model data D21, and the electrical shedding effect estimation calculation result data D23 is output.

Next, in processing step S34, one generator is selected in descending order of the power cut effect estimation calculation results. In processing step S35a, the transient stability calculation program P32 is executed in the state in which the electrical outage is selected, and the transient stability calculation result during the electrical outage of the power system 100 is output.

In processing step S35b, the grid stabilization confirmation program P35 is executed using this result and the threshold value data D17, and the grid stabilization confirmation result data D24 and the electrical control target selection result table D25 are output. Here, if it is determined that it is unstable, the process proceeds to processing step S33b, performs the electricity shedding effect estimation calculation after the electricity shedding machine is selected, and returns to processing step S34. Furthermore, if it is determined that the generator combination is stable, the generator combination is determined to be the target of power restriction in processing step S35d, and is stored in the power restriction target selection result table.
Note that when selecting electrical power reduction in the process of processing step S34, it is preferable to include a determination that the amount of power supply dropout is less than an allowable amount. If it is determined that the amount of power supply dropout is greater than or equal to the allowable power supply dropout amount, the supply and demand balance is maintained by simultaneously selecting load limiting targets.

In the overall flow shown in FIG. 11, the present invention is particularly characterized in that the learning process (processing step S31) of the calculation unit 41a is basically performed offline, and the electricity cutoff effect estimation model is created in advance. There is. This makes it possible to secure calculation time for preparing a plurality of highly accurate power reduction effect estimation models.

FIG. 13 shows a detailed flow of the electricity cut effect estimation model creation process in processing step S31. Note that the process shown in FIG. 13 may be performed by performing an offline test to confirm the fitting (confirm that the electric restriction effect estimation model is highly accurate, has versatility, and does not make a big mistake). It is preferable to issue an alarm if the result of this fitting confirmation is not favorable. This fitting confirmation is normally evaluated by inputting test data to the learning model, calculating an estimated value, and calculating the error between the estimated value and the true value calculated from the test data. It may be a simple method of keeping a combination of maximum and minimum values of and checking whether the test data is extrapolated to the learning data. This has the effect of reducing the calculation time for fitting confirmation. The pre-creation of the model will be explained in detail below.

Hereinafter, the procedure for creating an electricity restriction effect estimation model will be explained for each processing step in FIG. 13. First, in processing step S31a, the electrical braking transient stability calculation result data D11 and the learning setting value D12 are read into the memory.

In processing step S31b, the data is divided into model construction data (for learning) and model verification data (for testing). For example, measurement data obtained from the power system within a certain period of time is divided into appropriate time periods, and divided into data for model construction (for learning) and data for model verification (for testing).
In processing step S31c, standardization processing (scale conversion processing) is performed on the learning data as preprocessing. In the standardization process, as shown in equation (3), each variable is subtracted by its average value and then divided by its standard deviation to perform scale conversion so that the average becomes 0 and the variance becomes 1. Note that the test data is standardized by standardizing only the learning data that can be obtained offline and using the statistical values at that time to standardize the test data. By standardizing in this way, for example, the effective power P of the learning data can be changed to p. u. Value: 0.0 to 1.1 p. u. The generator internal phase difference angle δSG is set within a range of deg. When the value ranges from 0 to 180 degrees, the scale is different. Therefore, standardization has the effect of preventing the range of possible values for each data type (scale is different, and it is impossible to compare the influence of each coefficient of each explanatory variable on the objective variable. Depending on the regression model, it has the effect of increasing learning speed and learning accuracy.

Here, xia is the explanatory variable after standardization, xib is the explanatory variable before standardization, xic is the average value of explanatory variable xib before standardization, σx is the standard deviation of explanatory variable xib before standardization, and n is the number of explanatory variables. It is.

Further, as pre-processing, data reduction processing is executed, which includes one or more of processing such as thinning data to a predetermined amount of data and reducing data by focusing only on changing points. Note that the order of the division processing in processing step S31b and the preprocessing in processing step S31c may be reversed.

In processing step S31d, machine learning using a regression model is performed using the preprocessed learning data. By using a regression model, it becomes clear which explanatory variables have large coefficients and are important, which improves readability for operators and planners. Additionally, there is an effect that can be explained. Furthermore, variable selection for sparse regression, which will be described later, has the effect of making it possible to understand the importance of explanatory variables.

As a processing result of processing step S31d, the electricity restriction effect estimation model finally obtained is stored in the electricity reduction effect estimation model database DB21, but when it is obtained by sparse regression, this model format can be called a sparse model. , and is similarly stored in the electricity restriction effect estimation model database DB21.

Here, an image of the learning data is shown using FIGS. 14a to 14e. First, an image of deceleration energy estimation using machine learning will be explained with reference to FIG. 14a. In these figures, the vertical and horizontal axes represent a P-δ curve in a plane using the active power output _PSGj of the synchronous generator SGj and the internal phase difference angle δSGj. Here, in this specification, the suffix of the synchronous generator SG that is subject to power shedding is i and is expressed as SGi, and the suffix of the synchronous generator SG that tends to accelerate other than the shedding machine during SGi power shedding is j and is expressed as SGj. It is expressed as.

As shown in the figure, the active power output P _SGj and the internal phase difference angle δSGj are determined at the operating point, during fault, before fault removal, before electrical shedding, before electrical shedding, and after electrical shedding, as the system status changes. It is on the P-δ curve of Therefore, the plot points of the active power output P and the internal phase difference angle δ from the operating point to just before the electrical outage are adopted as explanatory variables in the learning data, and the deceleration energy DEij ( T) is adopted as the objective variable, and learning is performed using the following regression model.

Multiple regression analysis (linear regression model, linear multiple regression model) is a regression equation in which one objective variable is expressed by multiple explanatory variables, as shown in equation (4). Note that simple regression analysis is a regression equation in which one objective variable is expressed by one explanatory variable. Here, f(x) is an objective variable, β0 is a constant (intercept), βp is a partial regression coefficient, and x1 to xp are explanatory variables.

The learning data (set of explanatory variables and objective variables) is substituted into the above equation (4), and constants and partial regression coefficients are calculated by the method of least squares. Here, the explanatory variables are all plot points of the active power output _PSGj and the internal phase difference angle δSGj from the initial operating point to the operating point immediately before the electrical outage, and the objective variable is the operating point from immediately after the electrical outage to the operating point after a certain period of time. Let this be the deceleration energy calculated from .

Note that the learning model is created for each power cut case (power cut of the synchronous generator SG or renewable energy power source RES) using explanatory variables and objective variables for each synchronous generator SG. By estimating the power shedding effect using this learning model, it is possible to compare the power shedding effects of the synchronous generator SG and the renewable energy power generation RES using the same index.

Here, the challenge when learning using a regression model is Higo, Tsurumi, and Yumoto: ``Short-term future solar radiation prediction based on surface meteorological observation data - a prediction method using pyranometer data from multiple locations.'' ”, Central Research Institute of Electric Power Industry Report, R14019 (2015/7) p. 2 and Ohno: "Introduction to Multivariate Analysis", Doyukan (1998/10) p. 77, overfitting and multicollinearity.

Overfitting refers to a tendency for prediction accuracy to decrease when there are too many explanatory variables in a prediction formula relative to the number of learning data. The least squares method used in linear regression determines regression coefficients to best fit a limited number of data. Therefore, even explanatory variables such as noise that have no correlation with the target variable may appear to have a correlation with the target variable within limited data. In this case, the prediction formula is adapted to a location where the explanatory variable and objective variable, such as noise, happen to behave in the same way, resulting in a decrease in prediction accuracy. It can be seen that when overfitting occurs, the accuracy is very high for the training data (the coefficient of determination ^R2 is close to 1), but the error for new data increases. Even in linear multiple regression, problems of overfitting and multicollinearity may reduce the versatility of the model for new data.

Furthermore, multicollinearity refers to a tendency for prediction accuracy to decrease when explanatory variables have a high correlation. Mathematically, when the correlation between variables is high, the value of the determinant required in the inverse matrix calculation when determining the coefficients (partial regression coefficients) of a multiple regression equation approaches 0 and becomes unstable. As a result, one of the coefficients of explanatory variables that are highly correlated will have a large positive value, and the other will have a negative and large value.

FIG. 14b is a diagram for explaining an application image of sparse regression and the like to improve accuracy. Regarding the problem of overfitting and multicollinearity, for example, as _shown in FIG. ×2, overfitting and multicollinearity occur, whereas, as shown in FIG. 14b, for example, as learning data, the active power output P _SGj and the The problem can be avoided by setting the changing points of the phase difference angle δSGj to 6 points×2. Note that the objective variable remains the deceleration energy calculated from immediately after the electrical braking to the operating point after a certain period of time. In FIG. 14b, the reduced explanatory variables are indicated by black circles (●), but various variations of the explanatory variables can be applied. As a method for reducing explanatory variables as a countermeasure against overfitting and multicollinearity, in addition to the above-mentioned method of extracting only changing points, a variable reduction method, a variable increase/subtraction method, etc. may be used.

In addition to reducing explanatory variables, regularization can be cited as a countermeasure against overfitting and multicollinearity. By using this regularization, it is possible to obtain regression coefficients that avoid excessive complexity of the model by providing a penalty for increases in complexity during model learning.

The sparse regression method using the regularization method will be explained below. Lasso (Least Absolute Shrinkage and Selection Operator, Lasso) proposed by Tibshirani in 1996 is a typical method of the L1 regularization method in sparse modeling. Lasso is a method of estimating parameters by minimizing a regularization loss function obtained by adding a regularization term based on the parameter L1 norm to the loss function of a regression model, and can perform variable selection as well as stabilization of estimation. In addition, as derivatives of Lasso, there are Fused Lasso, Group Lasso, overlapping group Lasso, and OSCAR (Octagonal Shrinkage and Clustering Algorithm for Regression). n), Cluster Lasso, and Adaptive Lasso, but in the present invention, the term Lasso includes derivatives of these. .

There are several types of regularization methods, and regression models called Ridge, Lasso, and Elastic Net are often used. Ridge regression uses L2 regularization, and Lasso regression uses L1 regularization.

Here, the least squares method in the multiple regression analysis described above can be expressed as in the following equation (5). Note that here, yi is the objective variable of the learning data, and xi is the explanatory variable of the learning data.

Original author: Trevor Hastie, Robert Tibshirani, Jerome Friedman, supervised translation: Sugiyama, Ide, Kamijima, Kurita, Maeda: "Basics of statistical learning - data mining, inference, prediction -", Kyoritsu Shuppan (2014/6), pp. As shown in 76 to 86, Ridge regression can be expressed as in equation (6) by adding a penalty term (regularization term) to equation (5). Here, α is a parameter (hyperparameter) that controls the degree of reduction, and as the hyperparameter α increases, the degree of reduction also increases. Also, the coefficients are scaled toward zero with respect to each other.

Further, in Lasso regression, the penalty term in equation (6) is replaced with L1 regularization, resulting in equation (7). Here, since the second term constraint causes a nonlinear solution to yi, the solution is obtained by solving a quadratic programming problem.

Finally, Elastic Net regression combines the above two to become equation (8). Here, γ is the ratio of the L1 penalty term and the L2 penalty term. For example, set γ=0.5.

The characteristics of each regression model are summarized as follows. Although three reduced regression models are shown here, other reduced regression models may be used. First, Ridge regression has the effect of reducing regression coefficients, avoiding overfitting and multicollinearity, and preventing large deviations. By reducing the regression coefficient of the explanatory variable to zero, there is an effect of selecting variables, and Elastic Net regression has the effect of combining Ridge regression and Lasso regression.

In addition, perform K-fold cross-validation (K-fold cross-validation (cross-validation)) on the training data, and set K = 5 and set the sparse regression hyperparameter α to, for example, 10 ^-50 to 10 ⁺⁵⁰ . Searching (grid search) in increments of ¹⁰¹ and determining the result has the effect of generating a versatile model. Here, K-fold cross-validation is performed, for example, in the original author: Trevor Hastie, Robert Tibshirani, Jerome Friedman, supervised translation: Sugiyama, Ide, Kamishima, Kurita, Maeda: "Fundamentals of Statistical Learning - Data Mining, Inference, and Prediction". , Kyoritsu Shuppan (2014/6), pp. The methods shown in 227-279 are used.

FIG. 15 is a diagram showing an image of the hyperparameter α. This diagram shows the effect on other generators when power reduction processing is performed on individual generators (synchronous generator SG and renewable energy generation RES) for each type of assumed failure (failure point and failure mode). The relationship is expressed as an index using the hyperparameter α.

Note that FIG. 16 shows an image of changes in the regression coefficients of each variable with respect to changes in the hyperparameter α during this search. Figure 16 shows the regression coefficient of each variable on the vertical axis and the hyperparameter α on the horizontal axis in logarithmic notation. By displaying this, it is possible to confirm whether the coefficient has been sufficiently reduced in the hyperparameter α calculated by cross validation. Here, the hyperparameter α obtained by 5-fold cross validation and 10-fold cross validation may be different. In that case, in the accuracy evaluation, it is determined which hyperparameter α is better, and the hyperparameter α with higher accuracy is determined. The parameter α can be selected. In addition, if the hyperparameter α obtained as a result of cross-validation is stuck at the upper and lower limits of the search range 10 ^-50 to 10 ⁺⁵⁰ , it is possible to make appropriate adjustments by issuing an alert or expanding the search range. re-search for the hyperparameter α. Thereby, an appropriate hyperparameter α can be selected, which has the effect of improving the accuracy of the learning model. Note that cross-validation refers to cross-validation, and is preferably set at values such as 5-fold or 10-fold, and used when, for example, it is desired to automatically determine the hyperparameter α. In addition, in Fig. 16, the symbols t = 1 - (before failure, before circuit change), t = 1 + (after failure, after circuit change), t for the active power P of the synchronous generator and the internal phase difference angle δ (expressed as DEL) are shown. = 1.05 + (before fault removal, before circuit change), t = 1.05 - (after fault removal, after circuit change), t = 1.10 (midpoint between after fault removal and before power cut, no circuit change) ), t = 1.15 - (before electrical control, before circuit change) are expressed as each variable.

Note that various types of objective variables can be adopted. In place of the deceleration energy before and after electrical braking in FIG. 14a, FIG. 14c is a diagram showing an image when the target variable is the deceleration energy deviation ΔDEij(T) before and after electrical braking. This has the effect that the effect can be seen by focusing only on the amount of change in deceleration energy.

Furthermore, as shown in FIG. 14d, the maximum value δSGjmax of the internal phase difference angle after electrical braking may be used as the objective variable, or as shown in FIG. 14e, the maximum value deviation ΔδSGjmax of the internal phase difference angle after electrical braking may be used as the objective variable. . By doing so, it becomes possible to select electric control using the magnitude of δ or Δδ as an index rather than deceleration energy, which has the effect of making it easier for the operator to understand. In any case, at the time of learning, it is sufficient to use the parameters immediately before the power outage and the parameters after the power outage.

Returning to the process in FIG. 13, in process step S31e, accuracy evaluation is performed on the regression model obtained by performing machine learning. If the model accuracy is less than the threshold in processing step S31f, correction by preprocessing is performed in processing step S31g. Basically, it does things like reduce data. Furthermore, if the model accuracy is equal to or greater than the threshold value, the process is ended and the process proceeds to step S32a. Note that the threshold value is illustrated in FIG. 10, and threshold value data D17b stored in the threshold value database DB17 is used to evaluate the electricity restriction effect estimation model created in the calculation unit 41a.

Regarding the processing in FIG. 13, the present invention is equipped with a learning data sufficiency confirmation function that confirms that the learning data is sufficiently wide so that the actual non-electrostatic transient stability calculation results are not extrapolated. desirable. As a specific example, after the data reading step S31a, the electrical shedding effect estimation model creation unit 31 separately calculates waveforms such as transient stability calculation results P and δ in severe peak/off-peak cross sections, Compare these data with the read data to check if there is anything outside the data. If it is sufficient, proceed to data division. If not, use the harsh method described above. It is preferable to expand the sufficiency of the learning data by adding waveforms such as the transient stability calculation results P and δ in the peak-off-peak cross section to the learning data.

If these things are not done, data will be extrapolated to the model, resulting in a very large error, so by taking the above steps, you can create a model that does not undergo extrapolation, and improve the accuracy of the model. be able to.

The above is a point to keep in mind when creating a model in the calculation unit 41a, but the calculation unit 41b also checks whether the transient stability calculation result data D22 is extrapolated to the electric shedding effect estimation model. It is preferable to provide a checking mechanism before the estimation calculation unit 33 to determine whether or not this electricity restriction effect estimation model can be used.

In this case, it is necessary to evaluate whether P and δ of the transient stability calculation result data D22 are between the maximum and minimum values among the many P and δ of the transient stability calculation result data D11 during electrical braking. So, you can judge. This makes it possible to avoid making power control selection using a model that is extrapolated.

Note that when used for screening as in Example 2, which will be described later, when there is a risk of extrapolation, by not screening, it is possible to avoid a decrease in the selection accuracy of electric charge candidates. If there is a risk of extrapolation, an alert is issued and a different index, such as the sensitivity of the electric control, is used to select the electric control, thereby making it possible to select the electric control.

FIG. 17 shows a detailed flow for newly calculating these data when the transient stability calculation result data D11 during power reduction in processing step S31a of FIG. 13 does not exist or is insufficient. The learning and test data creation procedures will be explained for each processing step in FIG. 17.

In processing step S131 in FIG. 17, system equipment data, system measurement data, setting value data, and assumed failure data are read, or a power flow cross section or a state estimation cross section is read. At this time, various threshold values are also input.

In processing step S132a, a transient stability calculation program P32 is executed on the system equipment data D13, system measurement data D14, assumed failure data D15, or power flow cross-section data under the condition that no generator is cut off (no power cut), and the power The transient stability calculation result data D22 of the system 100 is output.

Specifically, in processing step S132a, a hypothetical failure is set, in processing step S132b, transient stability calculation is performed for the set failure, and in processing step S132c, transient stability calculation result data D22 and threshold data are calculated. Based on D17, it is determined whether the power system 100 is transiently stable or unstable. The stability is determined using, for example, the internal phase difference angle, frequency, voltage, etc. of the synchronous generator as indicators. If stable, the process returns to step S132a. Note that if all failures have been set, this processing flow ends. If it is unstable, the process moves to processing step S134, which will be described later.

In the processing step S134, the electrical restriction target selection program P34 is executed on the electrical restriction constraint data D16, the transient stability calculation result, and the threshold data. Next, in the processing step S135a, the transient stability The stability calculation program P32 is executed to output the transient stability calculation result during power outage of the power system 100. Note that the process in step S134 may be selected mechanically, and the indicator at this time may be determined based on whether or not electrical control is possible. You can also calculate only the missing parts. Conventionally, it seems that there is no data when electricity is cut off for renewable energy power sources, so it is better to calculate this in order to supplement it as learning data.

In processing step S135b, the grid stabilization confirmation program P35 is executed based on this result and the threshold data D17, and if it is determined that the system is unstable, the process returns to processing step S134, and in order to obtain the next electrical cutoff, The calculation for estimating the shedding effect is performed again, the next shedding machine is selected, and the transient stability calculation is performed, leading to processing step S135b.

If it is determined that the stability is stable, it is determined in processing step S132b whether all cross sections have been selected, and if not, the process returns to processing step S132a, the next cross section is set, and the process is continued. . If the selection of all other cross sections has been completed, the process proceeds to step S132c, and it is determined whether the selection of all possible failures has been completed. If the selection has not been completed, the process returns to step S32a. If the process has ended, the process advances to step S36.

By doing this, when the training data and test data are insufficient or the accuracy is insufficient, it is possible to increase the training and test data, which has the effect of improving the accuracy of the electricity restriction effect estimation model. be.

Here, an example of the creation result of the power cut effect estimation model will be explained using FIG. 18. In the table of FIG. 18, a multiple regression model is expressed as a coefficient for each explanatory variable of each generator for each failure point, mode, and power cutout machine. The figure shows an example where data is narrowed down to changing points in preprocessing. In addition, x in the table. The numbers indicated by xxx are not the same numbers. This point is the same for other charts.

The meaning of FIG. 18 will be explained using FIG. 19. FIG. 19 shows the change over time in the active power Pj of the power outage candidate SGj from before the failure occurs until after the power outage occurs. In Fig. 19, the symbols shown in the black circles are t = 1- (before failure, before circuit change), t = 1+ (after failure, after circuit change), t = 1.05- (before failure removal, before circuit change). ), t = 1.05 + (after fault removal, circuit change), t = 1.15 - (before power reduction, before circuit change), t = 1.15 + (after power reduction, circuit change), This is the time from before the failure occurs until the failure occurs, the fault is removed, and the power is cut off. Figure 18 shows the coefficients for each explanatory variable of each generator for each failure point, aspect, and power cut machine at this time. A regression model of is expressed. Here, although FIG. 19 only shows the effective power P meant in FIG. 18, the same applies to the internal phase difference angle δ. Furthermore, in FIG. 19, when the electric control start time is Ts, the calculation time width T represents the time for calculating the deceleration energy of the objective variable in the learning data.

Further, FIG. 20 shows an example in which a sparse regression (reduced regression) model is also applied. The meaning of FIG. 20 will be explained using FIG. 21. Although FIG. 19 shows only the time-series waveform of the active power P, FIG. 21 represents the P-δ curve of the active power P and the internal phase difference angle δ. FIG. 21 shows the active power P _SGj of the synchronous generator SGj and the internal phase difference of the synchronous generator SGj from before the failure occurs until the failure occurs and the power is cut off when the sparse regression (reduced regression) model is also applied. FIG. 6 is a diagram showing a comparison of changes in angle δSGj. In this figure, the symbols shown in the black circles are t=1- (before failure, before circuit change), t=1+ (after failure, after circuit change), t=1.05+ (before failure removal, before circuit change), t=1.05- (after fault removal, circuit change), t=1.15- (before power reduction, before circuit change), t=1.15+ (after power reduction, circuit change), are shown in Figure 19. is the time from before the occurrence of the same failure until the occurrence of the failure and the power outage, and for each failure point, mode, and power outage machine at this time, sparse regression (reduced regression) model is expressed. In addition, in Fig. 21, when the electric control start time is 1.15 s, the expression t = 1.15 + T after the calculation time width T is the time up to the time when the deceleration energy of the objective variable is calculated in the learning data. is expressed. Note that the models shown in FIGS. 18 and 20 are stored in the electricity cutoff effect estimation model database DB21 shown in FIG.

Here, in FIGS. 18 to 21, as shown in FIG. 16, t=1.10 (midpoint between after fault removal and before power cutoff, no circuit change) may be added, or t=1. You may subtract 15+ (after electrical installation, after circuit change). Note that increasing the number of appropriate explanatory variables has the effect of improving model accuracy, but increases the amount of calculations. In addition, increasing the number of inappropriate explanatory variables may worsen model accuracy due to overfitting and multicollinearity, so this can be addressed by applying sparse regression or reducing the number of variables themselves. Thereby, model accuracy can be improved.

According to the example in which the sparse regression (reduced regression) model is also applied in Figure 20, for example, by using Lasso regression, some coefficients of the explanatory variables become zero (in the example in the figure, the column t = 1+), and the explanatory variables This has the effect of reducing overfitting and multicollinearity. This has the effect of improving model accuracy. Note that the same location does not necessarily become zero. The sparse regression shown in FIG. 20 is output to the outside. This has the effect of allowing planners and operators to understand which variables are important.

It is recommended that the power cut effect estimation model be created for each section (season, month, time of day, peak/off-peak, generator status, etc.). In that case, it is possible to switch by setting the date, power flow (for example, interconnection line power flow P with the main power system of the power supply system), and system information thresholds for each cross section so that switching can be performed. This has the effect of improving model accuracy.

Here, an example of the power cut effect estimation result will be explained using FIG. 22. The power shedding effect estimation results calculated from the power shedding effect estimation model of each generator for each power shedding machine at the relevant failure point/mode are shown, and the average value thereof is calculated and ranked in descending order.

Types of average include arithmetic mean (arithmetic mean), weighted mean, geometric mean (geometric mean), and harmonic mean. When using a weighted average, the sensitivity of each accelerating synchronous generator is weighted by, for example, the rated output of the accelerating synchronous generator, the internal phase difference angle at the end time of the transient stability calculation, or the amount of increase relative to its initial value.

The type of average can be set in advance, and the operator or planner of the power system can arbitrarily change the type via the input unit 12. By doing this, operators and planners can use more appropriate average types depending on the power system status and failures, leading to improved power grid selection accuracy.

At this time, by narrowing down the generators to only the synchronous generators SG that are accelerating, it is possible to see the shedding effect on the accelerating synchronous generators SG. Here, methods for determining accelerating synchronous generators include, for example, determining a synchronous generator whose speed deviation is always positive in the time range from the occurrence of a failure to the end time of transient stability calculation as an accelerating synchronous generator; There is a method in which a synchronous generator whose internal phase difference angle exceeds a threshold specified by threshold data D17 at the stability calculation end time is determined to be an accelerating synchronous generator. In this case, it is possible to use the maximum value instead of the average value, which has the advantage of being able to see the shedding effect on the maximum synchronous generator SG.

If there are multiple electric system candidates with the same rank, you can rank them by lowering the number of significant digits and comparing them, or you can number the electric system candidates and select them in descending order. Or, give priority to candidates with less past electrical work experience. This eliminates the possibility of having the same ranking, which not only improves the degree of freedom in settings for operators and planners, but also enables operators and planners to select electrical shedding equipment that meets their objectives.

Here, FIG. 23 shows an example of error distribution evaluation during learning of the electricity restriction estimation model. In this figure, the vertical axis shows the synchronous generators SG1, SG2, and SG3, and the horizontal axis shows the time when the synchronous generators SG1, SG2, and SG3 are cut off, and each synchronous generator SGi is shown in the intersecting area. The error distribution of each synchronous generator SGj at the time of control is displayed. It is desirable that the error distribution has a shape close to a normal distribution, and such a display format has the effect of making it easier to understand errors and the versatility of the model. Note that the notation between the same synchronous generators SG cannot be determined, so this field is left blank.

The notation in FIG. 23 makes it easy to check whether it is a general-purpose model and to evaluate whether it has a normal distribution. Note that if the coefficient of determination ^R2 is too good, such as 0.999, and overfitting or multicollinearity occurs, this error distribution will result in a large error. Although it is not shown in this figure, larger errors will result in a distribution that is significantly off.

Here, details of learning and an image of estimation calculation are shown using FIG. 24. In addition, in FIG. 24, the upper part shows the processing contents of the calculation unit 41a that executes offline machine learning to create the electricity restriction effect estimation model, and the lower part shows the processing contents of the calculation part 41b that executes online and calculates the electricity restriction effect estimation amount. Indicates the processing details.

First, the processing content of the calculation unit 41a that creates the electricity restriction effect estimation model will be explained. This content corresponds to the processing shown in FIG. 13. In other words, first, in processing step S31a in FIG. 13, the electrical braking transient stability calculation result data D11 and the learning setting value D12 are read into the memory. It has been described. That is, in order to calculate D11, first, a plurality of sets of power flow cross sections are calculated from the system equipment data D13 and the system measurement performance data D14 by state estimation power flow calculation. Note that the learning setting value D12 is used in machine learning processing described later.

On top of that, the transient stability calculation result data D11 during electrical outage is used to perform transient stability calculations at each time of no electrical power outage and during electrical outage, using electrical outage restriction data D16 and assumed failure data in addition to the power flow cross section. It is determined by This calculates the transient stability when it is assumed that electrical shedding will be performed in the event of multiple hypothetical failures.

Next, the division processing in processing step S31b and the preprocessing in processing step S31c are performed on the transient stability calculation result data D11 during electrical control. The data is divided into data for learning (for learning) D11A and data for model verification (for testing) D11B. Note that the electrical suppression effect amount when using the model construction data (for learning) D11A and the model verification data (for testing) D11B is obtained by calculation.

In processing step S31d, machine learning using a regression model is performed using the learning data D11A of the preprocessed transient stability calculation result data D11 during electrical outage. At this time, the main information included in the transient stability calculation result data D11 during power outage is the voltage V, internal operation angle δ, generator active power PG, generator reactive power QG, load active power PL, and load active power regarding the node. Power QL, active power P, reactive power Q, reactive power transmission loss Ploss, reactive power transmission loss Qloss for branches, generator terminal voltage Vt, internal phase difference δSG, generator active power PG, generator reactive power QG for the generator. , control system signals, limiter status, etc., and also includes the failure location, failure mode, failure duration, reclosing status, etc. as conditions for assumed failure. This input information is used as explanatory variable data and training data for machine learning. Through the learning process in processing step S31d, objective variable data and teacher data are output. Note that the learning setting value D12 is used in machine learning processing.

Thereafter, in processing step S31e, accuracy evaluation of the regression model obtained by performing machine learning is performed. At this time, the electricity reduction effect estimated amount obtained when applying the model verification data D11B to the model obtained by learning with the model construction data D11A, and the electricity reduction effect calculated from the model verification data (for testing) D11B are used. The amount (true value) will be compared. In addition, before this accuracy evaluation, at the time of model construction, the electric control effect estimation amount obtained when applying the model construction data D11A to the model obtained by learning with the model construction data D11A, and the model construction data D11A Basic evaluation may be performed by comparing the electric control effect amount (true value) calculated from .

Next, in processing step S31f, if the model accuracy is less than the threshold value, correction by preprocessing is performed in processing step S31g. Basically, it performs things like increasing and decreasing data (selecting variables). Specifically, variable reduction method, variable increase method, variable increase/decrease method, principal component analysis may be applied, or sparse regression (Ridge regression, Lasso regression, ElasticNet regression, etc.) may be applied. Note that when determining the hyperparameter α of sparse regression, it can be determined by performing a grid search using cross validation. In addition, when the model accuracy is equal to or higher than the threshold value, this is set as the finally obtained electricity restriction effect estimation model D21.

Next, the processing contents of the arithmetic unit 41b shown in the lower part of FIG. 24, which is executed online to calculate the estimated amount of power reduction effect, will be explained. Here, the transient stability calculation result data D22 during power reduction is obtained in processing step S32b shown in FIG. 11. The data used here is basically the same as when calculating D11, but the difference is that system information obtained online is used.

After that, in processing step S33, a power restriction effect estimation calculation is executed, and at this time, a previously determined power restriction effect estimation model is used.

Note that the teacher data, which is input explanatory variable data explained so far, was the generator active power PG and internal phase difference angle δ, but other data related to transient stability calculations, failure information, etc. were added to the explanatory variables. You can learn it. In this case, accuracy is expected to improve, but in this case, problems of overfitting and multicollinearity may occur, so reducing the number of explanatory variables or using sparse regression can be effective in avoiding this problem. be.

Next, an example of the result display by the display unit 11 will be explained using FIGS. 25 and 26. In FIG. 25, the display screen 90 is divided into a "search settings" field 91 and a "results" field 92, and electricity control target calculation result table data D25 and the like are displayed in the form of a table or a graph. In the "Search Settings" field 91, the time and assumed failure of the power flow cross section for which the result of selection of the electrical control equipment is to be displayed are set. In the "result" column 92, the shedding combination, the total amount of shedding, the estimated result of shedding effect, and the transient stabilization effect due to shedding are shown as internal phase difference angle waveforms of the synchronous generator before and after selection of the shedding machine. This allows power system operators and planners to easily confirm the target of power shedding for each power flow cross section and assumed failure, the total amount of power shedding, and the degree of stabilization effect.

In FIG. 26, the power restriction target calculation result table data D25 is displayed in a power system diagram. In FIG. 26, the display screen 90 is divided into a "search setting" field 91 and a "power system diagram" field 92, and the power control target calculation result table data D25 and the like are displayed in the form of a table or a graph. In addition to the assumed failure points, this system diagram also displays the generators connected to the grid, distinguishing them into those subject to power restriction and those subject to non-power restriction. In addition, acceleration synchronous generators, electrical control candidates, and the like may be shown on the system diagram. This has the advantage that power system operators and planners can easily grasp assumed failures, power cut targets, and the positional relationships of accelerating synchronous generators.

Regarding the second embodiment, differences and effects from the first embodiment will be explained using FIGS. 27, 28, and 29.

In the second embodiment, according to FIG. 27 illustrating a functional configuration example of the power system stabilization device 10 according to the second embodiment of the present invention, the power system before the power control target selection unit 34d of the power system stabilization device 10 A charge reduction candidate screening unit 37 is provided to screen charge reduction candidates, and a charge reduction effect estimation model is used for this screening.

Further, according to FIG. 28 showing an example of the hardware configuration of the power system stabilizing device 10 according to the second embodiment of the present invention, the power grid candidate screening result data D37 is stored in the power grid candidate screening result data database DB27.

Here, as shown in FIG. 29, the power restriction target selection unit 34d uses the power restriction data to select, for example, the top four vehicles from the ranking of the power restriction effect estimation results as the power restriction candidate screening result. This has the effect of reducing the number of charge control candidates and the number of charge selection calculations. Thereafter, the selection of electricity bill candidates may be performed using the method of Example 1 or using the sensitivity described in Patent Document 1. The second embodiment also has the advantage that past calculation results by the power system stabilization device 10 can be effectively utilized.

Note that the present invention is not limited to the embodiments described above, and includes various modifications. For example, the embodiments described above have been described in detail to explain the present invention in an easy-to-understand manner, and the present invention is not necessarily limited to having all the configurations described. Furthermore, it is possible to replace a part of the configuration of one embodiment with the configuration of another embodiment, and it is also possible to add the configuration of another embodiment to the configuration of one embodiment. Furthermore, it is possible to add, delete, or replace some of the configurations of each embodiment with other configurations. Further, each of the above-mentioned configurations, functions, processing units, etc. may be partially or entirely realized in hardware by, for example, designing an integrated circuit.

10: Power system stabilization devices 41a, 41b: Arithmetic unit 1, arithmetic unit 2
3: Control device 11: Display section 12: Input section 13: Communication section 14: Processor 15: Memory 31: Electricity control effect estimation model creation section 32: Transient stability calculation section 33: Electricity control effect estimation calculation section 34: Electricity control Target selection unit 35: Grid stabilization confirmation unit 36: Transmission unit 37: Electricity system candidate screening unit 43: Bus line D1: Input data D2: Output data D11: Transient stability calculation result data during electric system system D12: Learning setting value D13 : System equipment data D14: System measurement data D15: Supposed failure data D17: Threshold data D16: Electricity shedding constraint data D21: Electricity shedding effect estimation model data D22: Transient stability calculation result data D23: Electricity shedding effect estimation calculation result data D24 : Grid stabilization confirmation result data D25: Power reduction target selection result data D27: Power reduction candidate screening result data DB11: Transient stability calculation result database during power reduction DB12: Learning setting value base DB13: System equipment database DB14: System measurement database DB15: Expected failure database DB17: Threshold database DB16: Electricity shedding constraint database DB1: Program database DB21: Electricity shedding effect estimation model database DB22: Transient stability calculation result database DB23: Electricity shedding effect estimation calculation result database DB24: Grid stabilization confirmation Result database DB25: Power restriction target selection result database DB27: Power restriction candidate screening result database 30a, 30b: Measuring device 100: Power system 110a to 110c: Synchronous generator or renewable energy power source 120a to 120c, 121a to 121c: Node ( bus line)
130a-130c: Transformer 140a-140c: Branch (line)
200: Communication network

Claims (26)

An electric power system stabilization device that determines a target for power outage to perform stabilization control in the event of a failure in the electric power system, according to the stability at the time of an assumed failure in the electric power system including a synchronous generator and a renewable energy power source,
a power shedding effect estimation model creation unit that creates a power shedding effect estimation model using a regression formula and machine learning based on the transient stability calculation results when performing power shedding when the assumed failure occurs in the power system;
a power shedding effect estimation calculation unit that estimates a power shedding effect due to power shedding of one or more of a synchronous generator and a renewable energy power source based on the power shedding effect estimation model and transient stability calculation results;
Based on the magnitude of the estimated calculation result of the power shedding effect, one of the power shedding candidates, a synchronous generator and a renewable energy power source, is selected for power shedding, and the power shedding effect is maintained until the transient stability of the power system is stabilized. an electricity restriction target selection unit that selects electricity restriction candidates as electricity restriction targets based on the estimated calculation results;
A power system stabilizing device comprising: an output section that outputs the results of each section. The power system stabilization device according to claim 1,
The electrical shedding effect estimation model creation unit uses various electrical quantities from the operating point to immediately before the electrical shedding as explanatory variables, based on the transient stability calculation results during electrical shedding and the learning setting values, and calculates the amount of electricity for a certain period of time immediately after the electrical shedding as explanatory variables. A power system stabilization device characterized by performing machine learning using a regression model using various electrical quantities up to the end as objective variables to create a power cut effect estimation model. The power system stabilization device according to claim 2,
The electrical quantities include the internal phase difference angle, active power output, reactive power output, speed deviation, terminal voltage, node voltage, phase angle, branch active power output, reactive power output, and renewable energy power source of the synchronous generator. A power system stabilizing device characterized by having two or more of an active power output, a reactive power output, a speed deviation, and a terminal voltage. The power system stabilization device according to claim 2,
The electrical quantities include voltage, internal operating angle, generator active power, generator reactive power, load active power, load active power for the node, active power, reactive power, reactive power transmission loss, reactive power transmission loss for the branch, A power system stabilizing device characterized in that it includes a generator terminal voltage, an internal phase difference angle, a generator active power, a generator reactive power, a control system signal, and a limiter status regarding the generator. The power system stabilization device according to claim 1,
The power system stabilizing device, wherein the regression equation is linear multiple regression, or sparse regression such as Ridge regression, Lasso regression, or ElasticNet regression. The power system stabilization device according to claim 1,
An electric power system stabilization device characterized in that the accuracy of the electricity shedding effect estimation model is confirmed using test data for the electricity shedding effect estimation model. The power system stabilization device according to claim 1,
An electric power system stabilization device characterized in that, when the accuracy of the power cut effect estimation model is insufficient, preprocessing is performed to reduce learning data to only change point data. The power system stabilization device according to claim 1,
The power shedding effect estimation model is based on the calculation results of transient stability during power shedding and learning setting values, and calculates what is required for the accelerating tendency synchronous generator with respect to the failure point, failure mode, and shedding machine. Characteristic power system stabilization device. The power system stabilization device according to claim 1,
The machine learning uses plot points of the internal phase difference angle and active power output from the operating point of the synchronous generator to just before the power cut as explanatory variables, and as the objective variable, a certain period of time after the power cut of the synchronous generator, or It is characterized by the deceleration energy up to the timing when the active power reaches its maximum value, the deceleration energy deviation before and after the electrical shedding, the peak value of the generator internal phase difference angle, or the peak value deviation of the generator internal phase difference angle before and after the electrical shedding. power system stabilization device. The power system stabilization device according to claim 1,
The accuracy of the shedding effect estimation model is determined by the coefficient of determination, the maximum and minimum values of the deviation of the error of the test results (estimated deceleration energy value) using the learning model, the maximum and minimum values of the absolute value of the deviation of the error, A power system stabilizing device characterized in that evaluation is performed using one or more root mean square errors RMSE. The power system stabilization device according to claim 1,
The power restriction target selection unit uses the power restriction effect estimation calculation result and the power restriction restriction data to select one or more of an arithmetic mean, a weighted average, a geometric mean, a harmonic mean, or a maximum value of the power restriction effect estimation calculation. An electric power system stabilizing device characterized in that the devices are selected one by one using the following. The power system stabilization device according to claim 1,
comprising an electric power restriction candidate screening unit that screens electric power restriction candidates based on the magnitude of the estimated calculation result of the electric power restriction effect;
The power cut target selection unit selects one of the synchronous generator and renewable energy power source from the screening results of the power cut candidates, and estimates the power cut effect until the transient stability of the power system is stabilized. An electric power system stabilizing device that selects power control candidates as power control targets based on calculation results, sensitivity, generator internal phase difference angle, and any index related to generator energy. The power system stabilization device according to claim 1,
The output unit includes system equipment data, system measurement data, assumed failure data, threshold data, power shedding constraint data, power shedding effect estimation model data, transient stability calculation result data, pre-shutdown effect estimation calculation result data, and system stability. 1. A power system stabilization device, characterized in that it displays at least one of the following: power restriction target selection result data, power restriction target selection result data, sensitivity calculation result data, power restriction candidate screening result data, and power restriction target calculation result data. The power system stabilization device according to claim 1,
The power system stabilizing device is configured to include a control device that stores the power restriction target given by the output unit and controls the stored power restriction target when a failure occurs in the power system. Characteristic power system stabilization device. The power system stabilization device according to claim 1,
a first processor that executes the electricity restriction effect estimation model creation section; a second processor that executes the electricity restriction effect estimation calculation section, the electricity restriction target selection section, and the output section; A power system stabilizing device characterized in that a first processor is executing processing in a previous stage. The power system stabilization device according to claim 1,
An electric power system characterized in that a control cycle of a processor that executes the power cut effect estimation model creation section and a control cycle of a processor that executes the power cut effect estimation calculation section, power cut target selection section, and output section are different. Stabilizer. The power system stabilization device according to claim 1,
The electric power restriction effect estimation model creation unit performs transient stability calculation to obtain transient stability calculation result data during electric power restriction, and uses the transient stability calculation result data during electric power restriction as data for model construction in machine learning and model verification. It is characterized by creating a model for estimating the power shedding effect by storing the data separately for use and evaluating the accuracy of the power shedding effect when machine learning is performed using the data for model construction and the data for model verification. power system stabilization device. The power system stabilization device according to claim 1,
The power shedding effect estimation model creation unit uses machine learning to create a power shedding effect estimation model using a regression equation and machine learning based on the transient stability calculation results when power shedding is performed at the time of the assumed failure in the power system. A power system stabilizing device characterized by comprising a learning data sufficiency confirmation function that confirms that the learning data is sufficiently wide data so that the actual powerless control transient stability calculation results are not extrapolated. The power system stabilization device according to claim 1,
The electricity shedding effect estimation calculation unit estimates the electricity shedding effect due to electricity shedding of one or more of the synchronous generator and the renewable energy power source based on the electricity shedding effect estimation model and the transient stability calculation results. A power system stabilizing device characterized by checking that transient stability calculation result data is not extrapolated to a model before a power cut effect estimation calculation is performed. The power system stabilization device according to claim 1,
The output unit is characterized in that the output unit displays the electric power shedding effect estimation result by numerically writing it on a matrix-like screen on which a plurality of generators and a plurality of generators to be subjected to electric shedding control are arranged vertically and horizontally. conversion device. The power system stabilization device according to claim 20,
An electric power system stabilizing device characterized in that a ranking of recommending electric power control is given on the matrix screen. The power system stabilization device according to claim 1,
The output unit outputs the internal phase difference of the generator at each time from before the occurrence of the failure to the occurrence of the failure and the time when the power is cut off after the failure occurs, for each combination of the failure mode, the generator that performs the power cutoff, and the acceleration tendency synchronous generator. A power system stabilizing device characterized by outputting corner and grid effect estimation results in time series. The power system stabilization device according to any one of claims 20 to 22,
A power system stabilizing device, wherein the power cut effect estimation result is calculated using a sparse regression (reduced regression) model. The power system stabilization device according to claim 1,
The output section is a hyperlink that shows the regression coefficients of variables on one of the vertical and horizontal axes, and the relationship between the effects on other generators when power reduction processing is performed on individual generators on the other side for each type of assumed failure. A power system stabilizing device characterized by outputting a solution path diagram expressed using a parameter α. The power system stabilization device according to claim 7,
The power system stabilization device is characterized in that the preprocessing applies one or more of a variable reduction method, a variable increase method, a variable increase/decrease method, and principal component analysis. A power system stabilization method that determines a power grid target for implementing stabilization control when a failure occurs in a power system according to the stability at the time of an assumed failure in a power system including a synchronous generator and a renewable energy power source, the method comprising:
a step of creating a power shedding effect estimation model using a regression equation and machine learning based on the transient stability calculation results when performing power shedding at the time of the assumed failure in the power system;
A power shedding effect estimation calculation step of estimating a power shedding effect due to power shedding of one or more of a synchronous generator and a renewable energy power source based on the power shedding effect estimation model and the transient stability calculation results;
Based on the magnitude of the estimated calculation result of the power shedding effect, one of the power shedding candidates, a synchronous generator and a renewable energy power source, is selected for power shedding, and the power shedding effect is maintained until the transient stability of the power system is stabilized. an electricity restriction target selection step of selecting an electricity restriction candidate as an electricity restriction target based on the estimation calculation result;
A power system stabilization method comprising: an output step for outputting the results of each part.

Images