System for real-time control of nuclear power plants

Abstract

A real-time control system for a nuclear power plant according to an embodiment of the present invention is for controlling a nuclear power plant (1) including a reactor (2) comprising a core (3) and control rods (4), a pressurizer (5), a steam generator (6), and a coolant pump (7), and comprises: a state acquisition unit (20) that acquires state variables according to a preset reference period; an inference engine (30) that generates output data including control commands, derivation grounds, and safety margins using a large language model (31) according to the state variables acquired by the state acquisition unit (20); a physical verification unit (40) that verifies the physical validity of the control commands generated by the inference engine (30) in real time; a hybrid control arbitration unit (50) that determines a control mode according to the safety margin calculated by the inference engine (30) and the verification result of the physical verification unit (40) and finally confirms the control commands; and a command to control the nuclear power plant (1) according to the control mode determined by the hybrid control arbitration unit (50) and the control commands confirmed by the hybrid control arbitration unit (50). It is characterized by including a generating part (60).

Description

Technology Field

[0001] The present invention relates to a real-time control system for a nuclear power plant, and more specifically, to a real-time control system for a nuclear power plant capable of safely controlling the nuclear power plant in real time by setting the control mode to one of a model-driven mode, a hybrid mode, and a classical control mode according to a safety margin calculated based on the nucleation escape rate, the internal pressure of the reactor, and the temperature of the cladding material. Background Technology

[0002] A nuclear power plant refers to a power plant that generates electricity using nuclear energy. It generates steam by utilizing the energy released during nuclear decay or nuclear reactions, and then uses this steam to rotate a turbine to produce electricity.

[0003] Nuclear power plants are classified into light water reactors, heavy water reactors, and graphite reactors depending on the type of moderator. Light water reactors consist of reactor system facilities and power generation system facilities.

[0004] A nuclear reactor system facility comprises a reactor that converts a low-temperature coolant into a high-temperature coolant using heat generated by nuclear fission, and a steam generator that generates steam from the feedwater by exchanging heat between the high-temperature coolant converted by the reactor and the low-temperature feedwater.

[0005] The power generation system facility comprises a turbine that rotates by steam generated by an increase generator, a condenser that cools the steam passing through the turbine with cooling water and converts it into low-temperature feedwater, and a main feedwater pump that pumps the low-temperature feedwater converted in the condenser to a steam generator.

[0006] The control system of a nuclear power plant can detect multiple pieces of information through a sensor unit and continuously control and monitor multiple system facilities.

[0007] However, existing nuclear power plant control systems mainly use PID (Proportional-Integral-Derivative) control and MPC (Model Predictive Control), but there were problems in that it was difficult to respond to nonlinear dynamics, long-term phenomena (Xenon Swing), and transient states.

[0008] In addition, existing nuclear power plant control systems had the problem that it was difficult to utilize simulator codes and physical laws for nuclear power plants.

[0009] In addition, existing nuclear power plant control systems had the problem that not only were safety constraints not specified, but there was also a lack of a verification mechanism for the nuclear power plants. Prior art literature

[0010] (Patent Document 0001) KR 10-1625640 B1(Patent Document 0002) KR 10-1777179 B1 The problem to be solved

[0011] The present invention has been devised to solve the aforementioned problems, and the objective of the present invention is to provide a real-time control system for a nuclear power plant capable of safely controlling the nuclear power plant in real time by setting the control mode to one of a model-driven mode, a hybrid mode, and a classical control mode according to a safety margin calculated based on the nucleation escape rate, internal reactor pressure, and the temperature of the cladding material.

[0012] Furthermore, the objective of the present invention is to provide a real-time control system for a nuclear power plant capable of precisely controlling the plant in real time through a large language model learned of physical laws, safety guidelines, etc., rather than simple numerical comparison.

[0013] Furthermore, the objective of the present invention is to provide a real-time control system for a nuclear power plant that can ensure system reliability by multi-verifying data such as the law of conservation of energy, the law of conservation of mass, core reactivity, nucleation escape rate, internal core pressure, and cladding temperature for the nuclear power plant prior to performing a control process.

[0014] In addition, the objective of the present invention is to provide a real-time control system for a nuclear power plant that can prevent major accidents by detecting transient states in advance before an accident occurs, can precisely control the system by finely adjusting the output according to the situation, and can reduce human error by using a large language model learned from a large number of simulation data. means of solving the problem

[0015] In order to solve the technical problems described above, a real-time control system for a nuclear power plant according to an embodiment of the present invention is intended to control a nuclear power plant (1) including a reactor (2) comprising a core (3) and control rods (4), a pressurizer (5), a steam generator (6), and a coolant pump (7), and comprises: a state acquisition unit (20) that acquires state variables according to a preset reference cycle; an inference engine (30) that generates output data including control commands, derivation grounds, and safety margins using a large language model (31) according to the state variables acquired by the state acquisition unit (20); a physical verification unit (40) that verifies the physical validity of the control commands generated by the inference engine (30) in real time; a hybrid control arbitration unit (50) that determines a control mode according to the safety margin calculated by the inference engine (30) and the verification result of the physical verification unit (40) and finally confirms the control commands; and a nuclear power plant according to the control mode determined by the hybrid control arbitration unit (50) and the control commands confirmed by the hybrid control arbitration unit (50). It is characterized by including a command generation unit (60) that controls the power plant (1).

[0016] In addition, the above control mode is characterized by including a model-driven mode, a hybrid mode, and a classical control mode.

[0017] In addition, the inference engine (30) calculates a safety margin using (Equation 1), wherein (Equation 1) is set as MARGIN=MIN((DNBR-DNBRs) / DNBRs, (Ps-P) / Ps, (PCTs-PCT) / PCTs), wherein MARGIN is a safety margin, DNBR is a nucleation departure rate, DNBR is calculated as a value obtained by dividing the critical heat flux (Q''c) by the actual heat flux (Q''l), P is a core pressure, Ps is a reference core pressure, and PCT is the maximum temperature of the cladding material.

[0018] Additionally, the hybrid control arbitration unit (50) switches the control mode to a classical control mode when the safety margin is less than a preset first reference margin, the hybrid control arbitration unit (50) switches the control mode to a hybrid mode when the safety margin is greater than or equal to the first reference margin but less than a preset second reference margin, and the hybrid control arbitration unit (50) switches the control mode to a model-driven mode when the safety margin is greater than or equal to the second reference margin.

[0019] In addition, the first reference margin is set to 0.05 to 0.15, and the second reference margin is set to 0.25 to 0.35.

[0020] In addition, the physical verification unit (40) executes a plurality of verification modules in parallel to verify the law of conservation of energy, the law of conservation of mass, core reactivity, nucleus boiling escape rate, core pressure, and maximum temperature of the cladding material, respectively, and the hybrid control arbitration unit (50) is characterized by switching the control mode to a classical control mode regardless of the size of the safety margin when any of the plurality of verification modules is found to be unsatisfactory. Effects of the invention

[0021] In a real-time control system for a nuclear power plant according to one embodiment of the present invention, the control mode is set to one of a model-driven mode, a hybrid mode, and a classical control mode according to a safety margin calculated based on the nucleation escape rate, internal reactor pressure, and the temperature of the cladding material, thereby having the effect of safely controlling the nuclear power plant in real time.

[0022] In addition, the real-time control system for a nuclear power plant according to one embodiment of the present invention has the effect of enabling precise real-time control of the nuclear power plant through a large language model learned of physical laws, safety guidelines, etc., rather than simple numerical comparison.

[0023] In addition, the real-time control system for a nuclear power plant according to one embodiment of the present invention has the effect of ensuring the reliability of the system by multi-verifying data such as the law of conservation of energy, the law of conservation of mass, the reactivity of the core, the nucleation escape rate, the internal pressure of the core, and the temperature of the cladding material for the nuclear power plant before performing the control process.

[0024] In addition, the real-time control system for a nuclear power plant according to one embodiment of the present invention can prevent major accidents by detecting transient states in advance before an accident occurs, can precisely control the system by finely adjusting the output according to the situation, and can reduce human error by using a large language model that has learned a large amount of simulation data. Brief explanation of the drawing

[0025] Figure 1 is a diagram of a nuclear power plant. FIG. 2 is a configuration diagram of a real-time control system for a nuclear power plant according to an embodiment of the present invention. Figure 3 is a block diagram of the learning process of a large language model. Figure 4 is a block diagram of the detailed training process of a large language model. Figure 5 is a flowchart of the process for calculating the safety margin. Figure 6 is a flowchart of the control mode switching of the hybrid control arbitration unit. Figure 7 is a verification flowchart of the physical verification unit. Figure 8 is a verification flowchart of the physical verification unit. FIG. 9 is a flowchart of the process for calculating the final safety margin and determining the control mode. FIG. 10 is a flowchart of the process for calculating the target device health index. Figure 11 is a flowchart of the process for calculating the final loss function. FIG. 12 is a flowchart of the process for calculating the final safety margin. Specific details for implementing the invention

[0026] Hereinafter, in order to explain in detail enough for a person skilled in the art to easily implement the technical concept of the present invention, embodiments of the present invention will be described with reference to the attached drawings.

[0027] However, the following examples are merely illustrative to aid in understanding the present invention and do not reduce or limit the scope of the present invention.

[0028] Furthermore, the present invention may be implemented in various different forms and is not limited to the embodiments described herein.

[0029] Figure 1 is a diagram of the configuration of a nuclear power plant (1).

[0030] Referring to FIG. 1, the nuclear power plant (1) is configured to include a reactor (2), a pressurizer (5), a steam generator (6), and a coolant pump (7).

[0031] The reactor (2) converts the low-temperature coolant into the high-temperature coolant using heat generated through nuclear fission.

[0032] The pressurizer (5) is connected to the reactor (2) and maintains the high-temperature coolant discharged from the reactor (2) in a high-pressure state so that the high-temperature coolant does not boil.

[0033] The steam generator (6) is connected to the pressurizer (5) and heat exchanges the high-temperature coolant that has passed through the pressurizer (5) with the low-temperature feed water, thereby generating steam from the low-temperature feed water.

[0034] A coolant pump (7) is provided between the reactor (2) and the steam generator (6) to pump low-temperature coolant that has passed through the steam generator (6) into the reactor (2).

[0035] The reactor (2) is composed of a core (3) and control rods (4).

[0036] The core (3) is installed inside the reactor (2) and generates heat through nuclear fission.

[0037] The control rod (4) is inserted into the reactor (2) and provided in the core (3) to reduce the nuclear chain reaction occurring in the core (3) by absorbing neutrons.

[0038] Inside the core (3), a number of nuclear fuels that generate heat through nuclear fission are provided.

[0039] A cladding material surrounding the nuclear fuel is provided on the outer side of the nuclear fuel.

[0040] FIG. 2 is a configuration diagram of a real-time control system for a nuclear power plant according to an embodiment of the present invention.

[0041] Referring to FIG. 2, a real-time control system for a nuclear power plant according to one embodiment of the present invention is configured to further include a state acquisition unit (20), an inference engine (30), a physical verification unit (40), a hybrid control arbitration unit (50), and a command generation unit (60).

[0042] The state acquisition unit (20) acquires state variables according to a pre-set reference period.

[0043] The inference engine (30) generates output data including control commands, derivation grounds, and safety margins using a large language model (31) according to the state variables acquired by the state acquisition unit (20).

[0044] The physical verification unit (40) verifies the physical validity of the control command generated by the inference engine (30) in real time.

[0045] The hybrid control arbitration unit (50) determines the control mode based on the safety margin calculated by the inference engine (30) and the verification result of the physical verification unit (40), and finally confirms the control command.

[0046] The command generation unit (60) controls the nuclear power plant (1) according to the control mode determined by the hybrid control arbitration unit (50) and the control command confirmed by the hybrid control arbitration unit (50).

[0047] The basis for the above derivation serves as the reason for deriving the above control command and may be generated in accordance with the final safety analysis report and technical guidelines.

[0048] The above output data may further include statistical reliability for the above control command.

[0049] The above state variable data may include core state variables, primary system state variables, and secondary system state variables.

[0050] The above core state variables may include neutron flux, group 6 delayed neutron precursor, xenon concentration, iodine concentration, core reactivity, core internal pressure, cladding maximum temperature, and actual heat flux.

[0051] The above primary system state variables may include nuclear fuel temperature, coolant temperature, coolant level, coolant flow rate, coolant stock, high-temperature side temperature, low-temperature side temperature, pressurizer pressure, and pressurizer level.

[0052] The above secondary system state variables may include steam generator temperature, steam generator pressure, steam generator water level, main feedwater flow rate, turbine inlet pressure, and main steam flow rate.

[0053] The above actual heat flux refers to the nuclear fuel heat flux.

[0054] The above technical guidelines may include regulatory data.

[0055] The above regulatory data includes reference energy change rate, reference mass change rate, reference core reactivity, reference nucleate boiling departure rate, reference core pressure, reference cladding maximum temperature, and critical heat flux data.

[0056] Critical heat flux data refers to the heat flux data of the nuclear fuel at the time of nucleus boiling deviation.

[0057] Meanwhile, the above nuclear fuel can be formed in the shape of a rod.

[0058] The above coolant can be set to water.

[0059] The above nucleation refers to a phenomenon in which heat transfer is performed as bubbles form on the surface of the nuclear fuel due to heat generated from the nuclear fuel.

[0060] Xenon is a radioactive material produced as a byproduct of nuclear fission reactions, and it absorbs neutrons, reducing the efficiency of the control rod (4).

[0061] Iodine is one of the radioactive substances produced as a byproduct of nuclear fission reactions.

[0062] The above reference period can be set to 100ms.

[0063] Figure 3 is a block diagram of the learning process of a large language model (31).

[0064] Figure 4 is a block diagram of the detailed learning process of a large language model (31).

[0065] Meanwhile, the state acquisition unit (20) collects the source code of the simulator, the final safety analysis report and technical guidelines, and a number of driving scenarios.

[0066] The above driving scenario may include initial conditions, accident progression, control measures, and final state.

[0067] The inference engine (30) generates a dataset for training the large language model (31) according to the data acquired by the state acquisition unit (20).

[0068] The inference engine (30) trains the large language model (31) using the dataset.

[0069] The inference engine (30) can train the large language model (31) so that the large language model (31) can learn the physical laws, control logic, and safety constraints of the nuclear power plant (1).

[0070] Meanwhile, the above dataset may include multiple source codes, driving scenarios, initial conditions, final safety analysis reports, and technical guidelines.

[0071] The above control modes may include model-driven mode, hybrid mode, and classical control mode.

[0072] The above model-led mode refers to a mode led by the above large language model (31).

[0073] The above initial condition data may include initial condition data for nuclear fuel, initial condition data for xenon, and initial condition data for the system.

[0074] Figure 5 is a flowchart of the process for calculating the safety margin.

[0075] Referring to FIG. 5, the inference engine (30) calculates a safety margin using (Equation 1).

[0076] The above (Equation 1) is set as MARGIN=MIN((DNBR-DNBRs) / DNBRs, (Ps-P) / Ps, (PCTs-PCT) / PCTs).

[0077] Here, MARGIN is a safety margin, and the MIN function is a function that outputs the minimum value among multiple input values.

[0078] The above DNBR is the nucleation departure rate, and the above DNBR can be calculated as the value obtained by dividing the critical heat flux (Q''c) by the actual heat flux (Q''l).

[0079] The above DNBRs is the reference nucleation departure rate, which is set to 1.3.

[0080] The above P is the core pressure, the above Ps is the reference core pressure, and the above Ps can be set to 18 MPa.

[0081] The above PCT is the maximum temperature of the coating material, and the above PCTs are the maximum temperature of the reference coating material, which can be set to 1,204℃.

[0082] FIG. 6 is a flowchart of the control mode switching of the hybrid control arbitration unit (50).

[0083] Referring to FIG. 6, the hybrid control arbitration unit (50) switches the control mode to a classical control mode when the safety margin is less than a preset first reference margin.

[0084] The above hybrid control arbitration unit (50) switches the control mode to a hybrid mode when the safety margin is greater than or equal to the first reference margin and less than the preset second reference margin.

[0085] The above hybrid control arbitration unit (50) switches the control mode to a model-driven mode when the safety margin is greater than or equal to the second reference margin.

[0086] Meanwhile, the above hybrid mode means outputting the output values ​​of the large language model (31) and the classical control algorithm at a preset standard ratio.

[0087] The above standard ratio is set to 0.6 for the large language model (31) and 0.4 for the classical control algorithm.

[0088] The above first reference margin can be set to 0.05 to 0.15.

[0089] The above second reference margin can be set to 0.25 to 0.35.

[0090] Figure 7 is a verification flowchart of the physical verification unit (40).

[0091] Referring to FIG. 7, the physical verification unit (40) executes a plurality of verification modules in parallel to verify the law of conservation of energy, the law of conservation of mass, core reactivity, nucleus boiling escape rate, core pressure, and maximum temperature of the cladding material, respectively.

[0092] The hybrid control arbitration unit (50) can switch the control mode to model-driven mode when all verification modules pass.

[0093] The above hybrid control arbitration unit (50) can switch the control mode to a classical control mode regardless of the size of the safety margin when any of the plurality of verification modules fails.

[0094] Figure 8 is a verification flowchart of the physical verification unit (40).

[0095] Referring to FIG. 8, the plurality of verification modules may include an energy conservation verifier, a mass conservation verifier, a reactivity verifier, a DNBR verifier, a pressure verifier, and a temperature verifier.

[0096] The energy conservation verifier verifies that the error rate between the calculated value of the energy change rate and the reference value is less than 1%.

[0097] The energy conservation verifier calculates the energy conservation deviation according to (Equation 2-1) g_E=dE / dt-dE / dt_s-0.01.

[0098] The energy conservation verifier determines it as a pass if the above energy conservation deviation value is less than 0.

[0099] The above g_E is the energy conservation deviation value, the above dE / dt is the energy change rate, and the above dE / dt_s is the reference energy change rate.

[0100] The above energy change rate can be calculated as the difference between the core fission heat Q_in and the heat loss amount Q_out.

[0101] The mass conservation verifier verifies that the error rate between the calculated mass change rate value and the reference value is less than 0.5%.

[0102] The mass conservation verifier calculates the mass conservation deviation value according to (Equation 2-2) g_m=dm / dt-dm / dt_s-0.005.

[0103] The mass conservation verifier determines it to be a pass if the above mass conservation deviation value is less than 0.

[0104] The above g_m is the mass conservation deviation value, the above dm / dt is the mass change rate, and the above dm / dt_s is the reference mass change rate.

[0105] The above mass change rate can be calculated as the difference between the coolant inflow rate m_in and the coolant outflow rate m_out.

[0106] The reactivity verifier verifies whether the core reactivity is within the reference reactivity range.

[0107] The reactivity verifier calculates the core reactivity deviation value according to (Equation 2-3) g_rho=|rho|-rho_s.

[0108] The reactivity verifier determines it to be a pass if the above core reactivity deviation value is less than 0.

[0109] g_rho is the core reactivity deviation, rho is the core reactivity height, and rho_s is the reference core reactivity. rho_s can be set to 500.

[0110] The above core reactivity can be calculated by summing the reactivity according to control rod position, boron concentration in coolant, xenon concentration, system temperature, and bubble fraction.

[0111] The DNBR verifier verifies whether the DNBR exceeds the reference DNBR.

[0112] The DNBR verifier calculates the DNBR deviation value according to (Equation 2-4) g_DNBR=DNBRs-DNBR.

[0113] The DNBR verifier determines a pass if the DNBR deviation value is less than 0.

[0114] g_DNBR is the DNBR deviation value, and DNBR is the nucleation deviation rate.

[0115] DNBRs is the reference nucleate boiling deviation rate. DNBRs can be set to 1.3.

[0116] The pressure verifier verifies whether the core pressure is below the reference core pressure.

[0117] The pressure verifier calculates the coolant core pressure deviation according to (Equation 2-5) g_P=P-Ps.

[0118] The DNBR verifier determines it as a pass when the core pressure deviation is less than 0.

[0119] The above g_P is the core pressure deviation value, the above P is the core pressure, and the above Ps is the reference core pressure. The above P_s can be set to 18.

[0120] The temperature verifier verifies whether the maximum temperature of the covering material is below the maximum temperature of the reference covering material.

[0121] The temperature verifier calculates the maximum temperature deviation of the coating material according to (Equation 2-6) g_PCT=PCT-PCTs.

[0122] The temperature verifier determines it as a pass when the maximum temperature deviation of the coating material is less than 0.

[0123] The above g_PCT is the maximum temperature deviation of the covering material, the above PCT is the maximum temperature of the covering material, and the above PCTs is the reference maximum temperature of the covering material. The above PCTs can be set to 1,204℃.

[0124] Meanwhile, the temperature verifier can verify whether the nuclear fuel temperature is below the reference nuclear fuel temperature. The reference nuclear fuel temperature can be set to 2,800℃.

[0125] Figure 8 is a block diagram of the process for calculating the loss function.

[0126] Meanwhile, the inference engine (30) uses a loss function according to (Equation 3) so that the large language model (31) can be optimized.

[0127] The above (Equation 3) is set as L_total=α*L_control+β*L_physics+γ*L_safety.

[0128] The above L_total is a loss function, the above α is a weight related to control performance, the above β is a weight related to physical laws, and the above γ is a weight related to safety.

[0129] The above α can be 1, the above β can be 10, and the above γ can be 5.

[0130] The above weights are set to prioritize compliance with physical laws and ensuring safety.

[0131] Meanwhile, the inference engine (30) calculates the control loss according to (Equation 4) through a large language model (31).

[0132] The above (Equation 4) is L_control=∑[w_k*{(u_pred,k-u_target,k) / Scale_k} 2 It is set to ].

[0133] The above L_control is the control loss, and the above k is the control variable.

[0134] The above control variables may include control rod position, boric acid concentration, and steam flow rate.

[0135] The above u_pred,k is a predicted control value calculated through a large language model (31), the above u_target,k is a preset target control value, the above Scale_k is a preset coefficient based on the driving range for each control variable, and w_k is a preset weight based on the importance of each control variable.

[0136] The above inference engine (30) calculates the safety loss using (Equation 5) L_safety=EXP(-MARGIN).

[0137] The above L_safety is a safety loss, and the above safety loss is an item for preemptively securing safety before reaching an accident threshold.

[0138] The above EXP function is a function that outputs the input value raised to the power of e, and e is set to 2.718281.

[0139] The inference engine (30) calculates physical loss according to (equation 6).

[0140] The above (Equation 6) is set as L_physics=∑[MAX(0, g_i(x))].

[0141] The above L_physics is a physical loss, n is the number of constraints, and n can be set to 6.

[0142] The above g_i(x) is the deviation value for the i-th constraint. The above i can be set to a natural number.

[0143] The above g_i(x) can output 0 when the constraint is satisfied, and output a penalty proportional to the deviation value when the constraint is violated.

[0144] The above g_i(x) includes at least one of a number of constraints.

[0145] The above-mentioned constraints include energy conservation deviation, mass conservation deviation, core reactivity deviation, nucleate boiling deviation, core pressure deviation, and cladding maximum temperature deviation.

[0146] Meanwhile, when PCT is 1,150℃ and PCTs is 1,204℃, g_PCT=1150-1204=-54 and MAX(0, -54)=0, so it is a steady state with no physical loss.

[0147] When PCT is 1,210℃ and PCTs is 1,204℃, g_PCT=1210-1204=6 and MAX(0, 6)=6, it is a critical state where physical loss has occurred.

[0148] In a critical state, the large language model (31) can be forced to immediately relearn physical laws and safety regulations.

[0149] In the present invention, the weights of the large language model (31) can be updated so that the total loss L_total, which is summed by applying weights to the control loss, physical loss, and safety loss respectively, can be minimized.

[0150] Figure 9 is a flowchart of the process for calculating the final safety margin and determining the control mode.

[0151] Referring to FIG. 9, the state acquisition unit (20) acquires a state variable from the target device j.

[0152] The inference engine (30) generates a control command based on the state variable obtained by the state acquisition unit (20).

[0153] The physical verification unit (40) calculates a soundness index HI_j for the target device based on the state variables S_n,j obtained from the state acquisition unit (20).

[0154] The physical verification unit (40) verifies whether the control command generated by the inference engine (30) exists within a preset maximum response time τ.

[0155] The hybrid control arbitration unit (50) determines the control mode according to the final safety margin MARGIN_Total, which integrates the soundness index and physical safety limit calculated by the physical verification unit (40).

[0156] The above state variables may include vibration, current, and temperature data of the target device.

[0157] Figure 10 is a flowchart of the process for calculating the target device health index.

[0158] Referring to FIG. 10, the state acquisition unit (20) calculates the target device health index HI_j according to (Equation 7) based on the state variables S_n,j acquired from the target device j.

[0159] The above (Equation 7) is set as HI_j=EXP(-η_j*∑(λ_n*(S_n,j-S_n,ref) / (S_n,limit-S_n,j)*(1+dS_n,j / dt*Δα_n))).

[0160] The above η_j is an external environment severity factor, the above λ is a weight for a state variable, the above S_n,j is a state variable, and the above n is the number of state variables.

[0161] The above S_n,ref is a reference value for the state variable, and the above S_n,limit is a limit value for the state variable.

[0162] The above dS_n,j / dt is the rate of change of the state variable S_n,j over time, and the above Δα_n is the temporal degradation acceleration, which can be determined based on the mechanical properties of the target device and its past failure history.

[0163] The normalized deviation between the reference value S_n,ref and the limit value S_n,limit indicates the current degree of deterioration.

[0164] Here, by multiplying the hourly rate of change of state variable S_n,j dS_n,j / dt and the temporal degradation acceleration Δα_n, the rate of degradation, rather than a simple numerical deviation, can be reflected in the exponent.

[0165] In particular, the external environment severity factor η_j determines the attenuation rate of an exponential function, thereby forcing the integrity index to drop more sensitively to even minute changes in the case of target equipment in high-risk areas with high radiation doses, which can induce a conservative safety judgment.

[0166] Figure 11 is a flowchart of the process for calculating the final loss function.

[0167] Referring to FIG. 11, the inference engine (30) includes a large language model (31) trained such that the final loss function L_t is minimized according to (Equation 8).

[0168] The above (Equation 8) is set as L_t=α*L_control+β*L_physics+γ*L_safety+δ*L_maintenance.

[0169] The above L_maintenance is the maintenance loss, and the above δ is the weight for the maintenance loss.

[0170] The above large language model (31) is trained to suppress the change amount Δu_j of the control command using the target device as the target device health index HI_j is lower.

[0171] The above maintenance loss L_maintenance is calculated according to (Equation 9).

[0172] The above (Equation 9) is L_maintenance=∑(1 / HI_j*Δu_j 2 It is set to ).

[0173] The above Δu_j is the amount of change in the control command for the j-th target device.

[0174] The above maintenance loss L_maintenance has an inverse relationship with the health index HI_j of the target device.

[0175] If the health of a specific target device deteriorates and HI_j converges to 0, the maintenance loss L_maintenance value increases rapidly.

[0176] Therefore, the inference engine (30) minimizes the amount of control change Δu_j for the corresponding target device to reduce the total loss.

[0177] This generates an intelligent control effect in which a large language model (31) suppresses the use of the target device in order to preserve the remaining lifespan of the target device.

[0178] The above physical verification unit (40) calculates a response deviation value g_delay for the target device using (Equation 10) to verify the response delay caused by the performance degradation of the target device.

[0179] Through this, the physical verification unit (40) can verify the physical executableability of the control command generated by the inference engine (30).

[0180] The above (Equation 10) is set to g_delay=τ_actual-τ_limit.

[0181] The above g_delay is the response deviation value for the target device, the above τ_actual is the actual completion time of the operation of the target device after the control command, and the above τ_limit is the maximum response time allowed in the technical guidelines.

[0182] When the response deviation value g_delay exceeds 0 due to mechanical friction or aging of the target device, the physical verification unit (40) determines that the control command including the target device is rejected and blocks the output of the inference engine (30).

[0183] The hybrid control arbitration unit (50) can switch the control mode to a classical control mode when the response deviation value g_delay exceeds 0.

[0184] Figure 12 is a flowchart of the process for calculating the final safety margin.

[0185] Referring to FIG. 12, the hybrid control arbitration unit (50) uses the final safety margin MARGIN_Total derived by (Equation 11) as a measure for decision-making.

[0186] The hybrid control arbitration unit (50) determines the minimum value among the physical margin MARGIN_physics representing the physical critical state of the core (3) according to (Equation 11) and the value reflecting the control weight C_j for each target device in the integrity index HI_j for each target device as the final safety margin.

[0187] (Equation 11) is set to MARGIN_Total=MIN(MARGIN_physics, MIN_j(HI_j*C_j)).

[0188] Here, MARGIN_physics is a physical margin, representing the physical critical state of the core (3).

[0189] HI_j is the health index for each target device j, and C_j is the control weight for each target device.

[0190] Meanwhile, the physical margin can be set to MARGIN_physics=MIN[(CS_k,limit-CS_k) / (CS_k,limit-CS_k,ref)].

[0191] k can be an index for a specific core state variable. CS_k,limit is a limit value for a specific core state variable, CS_k is a specific core state variable, and CS_k,ref is a reference value for a specific core state variable.

[0192] The hybrid control arbitration unit (50) dynamically switches the control mode based on the size of the final safety margin MARGIN_Total.

[0193] When the final safety margin is greater than or equal to a preset second reference value, the hybrid control arbitration unit (50) switches to a model-driven mode in which the inference engine (30) takes the lead in optimizing driving efficiency.

[0194] When the final safety margin is greater than or equal to a preset first threshold and less than the second threshold, the hybrid control arbitration unit (50) switches to a hybrid mode in which the physical verification unit (40) limits the commands of the inference engine (30) in real time.

[0195] The hybrid control arbitration unit (50) blocks the inference engine (30) and switches to a classical control mode when the final safety margin is less than the first reference value or when the health index of a specific target device is less than or equal to a preset blocking threshold.

[0196] The hybrid control arbitration unit (50) forcibly switches to the classical control mode even if the physical state variable is within the normal range when the final safety margin is determined by a sharp drop in the target device health index (HIj).

[0197] This performs a Defense-in-Depth function that completely controls the uncertainty of AI control through physical soundness indicators.

[0198] Meanwhile, in the case of the reactor coolant pump (7), the vibration rate, which is one of the most critical defect indicators, can be set as a state variable S.

[0199] (Example 1) Reactor coolant pump with vibration velocity state variable of 3 mm / s

[0200] When the vibration speed reference value S_n,ref=1mm / s, the vibration speed limit value S_n,limit=10mm / s, the vibration speed state variable S_n,j=3mm / s, the external environment severity factor η_j=0.8, and the temporal deterioration acceleration (dS / dt*Δα)=0.5, the integrity index HI_j is calculated for the target equipment according to (Equation 7).

[0201] Here, the vibration speed limit is the vibration level at which operation of the reactor coolant pump (7) is stopped or immediate maintenance is required, and the vibration speed reference level is the normal operating vibration level when the reactor coolant pump (7) is newly installed.

[0202] The severity of the external environment can be set very sensitively, taking into account that the reactor coolant pump (7) is a high-speed rotating body. The deterioration acceleration means that the vibration speed increases steeply over time.

[0203] HI_j=EXP(-η_j*∑(λ_n*(S_n,j-S_n,ref) / (S_n,limit-S_n,j)*(1+dS_n,j / dt*Δα_n)))

[0204] HI_j=EXP(-0.8*((3-1) / (10-3)*(1+0.5)))=EXP(-0.342)=0.71

[0205] Therefore, the integrity index for the reactor coolant pump (7) is 0.71.

[0206] At this time, it is assumed that a control change amount Δu_j=20 units is input for the target device.

[0207] Here, Δu_j is a change in control for the target device, and represents the maintenance intensity for replacing the bearing of the reactor coolant pump (7) or performing alignment work.

[0208] Calculate the maintenance loss L_maintenance according to (Equation 9).

[0209] (Mathematical Equation 9) L_maintenance=∑(1 / HI_j*Δu_j 2 )

[0210] L_maintenance=1 / 0.71*20 2 =563

[0211] (Comparative Example 1) Reactor coolant pump with a vibration velocity of 8 mm / s

[0212] Since HI_j=EXP(-0.8*((8-1) / (10-8)*(1+0.5)))=0.015, L_maintenance=1 / 0.015*20 2 =26,666.

[0213] When the vibration velocity state variable increases from 3 to 8, the maintenance cost L increases exponentially.

[0214] This provides strong numerical grounds for operators to intervene immediately before the risk increases exponentially when vibration speeds rise, rather than attributing it to maintenance failures.

[0215] Maintenance losses L_maintenance include not only actual maintenance costs but also massive economic losses due to the shutdown of the power plant in the event of a reactor coolant pump (7) failure.

[0216] Since the lower the health index HI_j, the higher the probability of failure, 1 / HI_j acts as a penalty and can significantly increase losses.

[0217] Through the above formula, it can be simulated whether it is advantageous to perform small maintenance Δu frequently before the vibration velocity S_j increases in order to minimize maintenance loss L_maintenance, or to endure until just before the critical point and then perform large maintenance.

[0218] Through the above formula system, it can be seen that as the vibration speed of the reactor coolant pump (7) reaches a critical value, and as the rate of increase in vibration speed increases, the maintenance loss L_maintenance increases like a snowball.

[0219] At nuclear power plant sites, the optimal balance between preventive maintenance and condition maintenance can be found based on maintenance loss L_maintenance.

[0220] In a real-time control system for a nuclear power plant according to one embodiment of the present invention, the control mode is set to one of a model-driven mode, a hybrid mode, and a classical control mode according to a safety margin calculated based on the nuclear boiling escape rate, the internal pressure of the reactor, and the temperature of the cladding material, thereby having the effect of safely controlling the nuclear power plant (1) in real time.

[0221] In addition, the real-time control system for a nuclear power plant according to one embodiment of the present invention has the effect of being able to precisely control the nuclear power plant (1) in real time through a large language model (31) that has learned physical laws, safety guidelines, etc., rather than simple numerical comparison.

[0222] In addition, in a real-time control system for a nuclear power plant according to one embodiment of the present invention, before performing a control process, the law of conservation of energy, the law of conservation of mass, the reactivity of the core (3), the nucleation escape rate, the internal pressure of the core (3), and the temperature data of the cladding material are multi-verified for the nuclear power plant (1), thereby having the effect of ensuring the reliability of the system.

[0223] In addition, the real-time control system for a nuclear power plant according to one embodiment of the present invention can detect a transient state before an accident occurs to prevent a major accident, can finely adjust the output according to the situation to precisely control the system, and can reduce human error by using a large language model (31) that has learned a large amount of simulation data.

[0224] As described above, the main technical concept of the present invention is to provide a real-time control system for a nuclear power plant. The embodiments described above with reference to the drawings are merely one example, and the true scope of the present invention is based on the patent claims, but extends to various equivalent embodiments that may exist. Explanation of the symbols

[0225] 1: Nuclear power plant 2: Nuclear reactor 3: Core 4: Control rods 5: Pressurizer 6: Steam generator 7: Coolant pump 10: Nuclear Power Plant Real-time Control System 20: State Acquisition Section 30: Inference Engine 31: Large Language Models 40: Physical Verification Department 50: Hybrid control arbitration unit 60: Command generation section

Claims

Claim 1 A real-time control system for a nuclear power plant for controlling a nuclear power plant (1) including a reactor (2) comprising a core (3) and control rods (4), a pressurizer (5), a steam generator (6), and a coolant pump (7), comprising: a state acquisition unit (20) for acquiring state variables according to a preset reference period; an inference engine (30) for generating output data including control commands, derivation grounds, and safety margins using a large language model (31) according to the state variables acquired by the state acquisition unit (20); a physical verification unit (40) for verifying the physical validity of the control commands generated by the inference engine (30) in real time; and a hybrid control arbitration unit (50) for determining a control mode according to the safety margin calculated by the inference engine (30) and the verification result of the physical verification unit (40), and finally confirming the control commands. A nuclear power plant real-time control system comprising: a command generation unit (60) that controls the nuclear power plant (1) according to a control mode determined by the hybrid control arbitration unit (50) and a control command determined by the hybrid control arbitration unit (50); wherein the control mode includes a model-driven mode, a hybrid mode, and a classical control mode. Claim 2 delete Claim 3 In claim 1, the inference engine (30) calculates a safety margin using (Equation 1), wherein (Equation 1) is set as MARGIN=MIN((DNBR-DNBRs) / DNBRs, (Ps-P) / Ps, (PCTs-PCT) / PCTs), where MARGIN is a safety margin, DNBR is a nucleation departure rate, DNBR is calculated as a value obtained by dividing the critical heat flux (Q''c) by the actual heat flux (Q''l), P is a core pressure, Ps is a reference core pressure, and PCT is the maximum cladding temperature, in a real-time control system for a nuclear power plant. Claim 4 A real-time control system for a nuclear power plant according to claim 1, wherein the hybrid control arbitration unit (50) switches the control mode to a classical control mode when the safety margin is less than a preset first reference margin, the hybrid control arbitration unit (50) switches the control mode to a hybrid mode when the safety margin is greater than or equal to the first reference margin and less than a preset second reference margin, and the hybrid control arbitration unit (50) switches the control mode to a model-driven mode when the safety margin is greater than or equal to the second reference margin. Claim 5 A real-time control system for a nuclear power plant according to claim 4, wherein the first reference margin is set to 0.05 to 0.15 and the second reference margin is set to 0.25 to 0.

35. Claim 6 A real-time control system for a nuclear power plant according to claim 1, wherein the physical verification unit (40) executes a plurality of verification modules in parallel to verify the law of conservation of energy, the law of conservation of mass, core reactivity, nucleus boiling escape rate, core pressure, and maximum cladding temperature, respectively, and the hybrid control arbitration unit (50) switches the control mode to a classical control mode regardless of the size of the safety margin when any of the plurality of verification modules is found to be unsatisfactory.

Images