An earthquake early warning based emergency regulation method for a hydropower station

Abstract

This application discloses an emergency control method and system for hydropower stations based on earthquake early warning, relating to the field of hydropower engineering safety and intelligent control technology. The method includes: acquiring multi-source correlated data such as earthquake early warning data, hydropower station operation data, and historical archive data; predicting disaster effect parameters caused by future earthquakes using a disaster effect prediction model; combining the disaster effect parameters with the current state of the hydropower station and generating an initial response strategy using a deep reinforcement learning algorithm; then calculating revision coefficients using equipment feedback data to dynamically correct the initial strategy and form a final response strategy; controlling the hydropower station equipment according to the final strategy and recording data to optimize the model. This method can predict disaster effects in advance and combines multi-objective optimization with real-time equipment status for closed-loop control, effectively solving the problems of delayed emergency response, rigid strategies, and execution failure under extreme conditions in traditional emergency response systems, significantly improving the safety and reliability of hydropower station earthquake emergency response.

Description

Technical Field

[0001] This application relates to the field of hydropower engineering safety and intelligent control technology, specifically to an emergency control method and system for hydropower stations based on earthquake early warning. Background Technology

[0002] Hydropower stations, as vital water conservancy hubs and energy facilities, are directly related to the safety of life and property downstream. Earthquakes are one of the major natural disasters threatening the safety of hydropower stations. During an earthquake, strong ground movement can cause dam structure vibration, massive surges in the reservoir, and even gate deformation and blockage, generator shutdowns, and other accidents. Existing emergency response mechanisms for hydropower stations typically have the following problems: traditional methods rely heavily on sensors to trigger alarms after seismic waves arrive, and the control logic is mostly based on preset fixed rules (such as shutdown or full gate opening), lacking the ability to predict future disaster effects (such as surge height and dam stress evolution), often missing the optimal emergency response window. Existing emergency plans cannot be dynamically adjusted according to the current real-time water level, flow rate, and equipment health status of the hydropower station, which may lead to misoperation or overoperation. For example, blindly opening gates when the reservoir water level is low not only wastes water resources but may also cause unnecessary flooding downstream. Under extreme conditions such as earthquakes, parameters such as motor current and load of control equipment such as gate hoists will fluctuate drastically. Ideal control commands (such as the set opening speed) may fail to be executed due to equipment overload or aging, resulting in control failure. Summary of the Invention

[0003] The purpose of this application is to provide an emergency control method and system for hydropower stations based on earthquake early warning, which solves the problems of slow response, single strategy and lack of dynamic correction capability in the existing technology when dealing with earthquake disasters.

[0004] This application is achieved through the following technical solution:

[0005] The first aspect of this application provides a method for emergency control of hydropower stations based on earthquake early warning, including:

[0006] Multi-source correlation data is acquired and preprocessed to obtain preprocessed multi-source correlation data; the multi-source correlation data includes earthquake early warning data, hydropower station operation data, and historical archive data.

[0007] The pre-set disaster effect prediction model is invoked to analyze the pre-processed multi-source correlation data and predict the disaster effect parameters caused by earthquakes in the future.

[0008] Based on the disaster effect parameters and the current state of the hydropower station, a deep reinforcement learning algorithm is used to generate an initial response strategy for emergency control of the hydropower station.

[0009] The revision coefficient is calculated using the equipment feedback data of the controllable equipment of the hydropower station, and the initial treatment strategy is dynamically modified according to the revision coefficient to form the final treatment strategy;

[0010] The final disposal strategy is used to conduct emergency control of the controllable equipment in the hydropower station; the controllable equipment of the hydropower station includes at least the hydropower station gates;

[0011] It records all relevant data during the current emergency control process of the hydropower station, and optimizes the pre-set disaster effect prediction model and deep reinforcement learning algorithm based on all relevant data for subsequent emergency control of the hydropower station.

[0012] In one possible implementation, the earthquake early warning data includes earthquake intensity, epicenter location, and S-wave arrival time; the hydropower station operation data includes hydropower station location, real-time reservoir location, flow velocity, gate opening, hoist load, motor current, voltage, generator active power, generator reactive power, dam seepage pressure and / or displacement monitoring data; the historical archive data includes the most recent historical earthquake data, historical final disposal strategies, and historical hydropower station operation data.

[0013] In one possible implementation, a pre-set disaster effect prediction model is invoked to analyze the preprocessed multi-source correlation data to predict disaster effect parameters caused by earthquakes in the future, including:

[0014] The preprocessed multi-source correlated data are combined into a vector, and the vector is used as the input of a pre-set disaster effect prediction model to predict the surge height caused by the earthquake in the future, the stress at preset key points on the hydropower station dam, and the gate impact load, so as to obtain the disaster effect parameters caused by the earthquake in the future; wherein, the disaster effect prediction model is pre-set using LSTM.

[0015] In one possible implementation, the presetting method for the disaster effect prediction model includes:

[0016] The loss function is constructed as either the cross-entropy loss function or the root mean square loss function;

[0017] The disaster effect prediction model is trained and deployed with the goal of minimizing the loss function.

[0018] In one possible implementation, based on the disaster effect parameters and the current state of the hydropower station, a deep reinforcement learning algorithm is used to generate an initial emergency control strategy for the hydropower station, including:

[0019] The current reservoir water level, current inflow, current gate opening vector, and unit operating status vector of the hydropower station are obtained to obtain the current status of the hydropower station.

[0020] The disaster effect parameters and the current state of the hydropower station are constructed into a state vector, and a deep reinforcement learning algorithm is used to identify the state vector to determine the action vector; the action vector includes the opening / closing rate of the gate;

[0021] The aforementioned action vector is used as the initial response strategy for emergency control of the hydropower station.

[0022] In one possible implementation, the method further includes the following steps before using the deep reinforcement learning algorithm:

[0023] Construct a digital twin simulation model of the hydropower station and obtain historical status by running the digital twin simulation model of the hydropower station;

[0024] By acquiring historical actions, historical rewards, and the next historical moment's state from historical states, a historical experience pool is constructed.

[0025] The deep reinforcement learning algorithm is optimized using the experience data in the historical experience pool, enabling the digital twin simulation model of the hydropower station to accurately select action vectors.

[0026] In one possible implementation, the historical rewards are obtained through a reward function, and the reward function is set as follows:

[0027] ;

[0028] In the formula, For the reward function, As a weighting of safety indicators, Rewards for safety performance indicators As the weight of the flood discharge index, As a reward for flood discharge, As the weight of economic indicators, Rewards for economic indicators This is a penalty item.

[0029] In one possible implementation, the revision coefficient is calculated using equipment feedback data from the controllable equipment of the hydropower station, including:

[0030] Based on the equipment feedback data from the controllable equipment of the hydropower station, obtain the real-time status adjustment factor;

[0031] Obtain the equipment health index, historical average efficiency deviation under the same operating conditions, and seasonal correction coefficient for the controllable equipment of the hydropower station, and obtain the historical experience correction factor based on the equipment health index, historical average efficiency deviation under the same operating conditions, and seasonal correction coefficient.

[0032] The revision coefficient is obtained by weighting and summing the real-time state adjustment factor and the historical experience correction factor.

[0033] In one possible implementation, the initial handling strategy is dynamically modified according to the revision coefficient to form a final handling strategy, including: multiplying the initial handling strategy by the revision coefficient to obtain the final handling strategy.

[0034] Based on the same inventive concept, a second aspect of this application provides a hydropower station emergency control system based on earthquake early warning, comprising:

[0035] The data acquisition module is used to acquire multi-source correlated data and preprocess it to obtain preprocessed multi-source correlated data; the multi-source correlated data includes earthquake early warning data, hydropower station operation data and historical archive data.

[0036] The deep learning module is used to call a pre-set disaster effect prediction model to analyze the pre-processed multi-source correlation data and predict the disaster effect parameters caused by earthquakes in the future.

[0037] The strategy generation module is used to generate an initial emergency control strategy for the hydropower station based on the disaster effect parameters and the current state of the hydropower station using a deep reinforcement learning algorithm.

[0038] The strategy correction module is used to calculate the revision coefficient using the equipment feedback data of the controllable equipment of the hydropower station, and to dynamically correct the initial treatment strategy according to the revision coefficient to form the final treatment strategy.

[0039] An emergency control module is used to perform emergency control on the controllable equipment of the hydropower station in accordance with the final disposal strategy; the controllable equipment of the hydropower station includes at least the hydropower station gates;

[0040] The data augmentation module is used to record all relevant data during the current emergency control process of the hydropower station, and to optimize the pre-set disaster effect prediction model and deep reinforcement learning algorithm based on all relevant data for subsequent emergency control of the hydropower station.

[0041] Compared with the prior art, this application has the following advantages and beneficial effects:

[0042] This application provides a method and system for emergency control of hydropower stations based on earthquake early warning. The method includes: acquiring multi-source correlated data such as earthquake early warning data, hydropower station operation data, and historical archive data, and using a disaster effect prediction model to predict disaster effect parameters caused by future earthquakes; combining the disaster effect parameters with the current state of the hydropower station, and using a deep reinforcement learning algorithm to generate an initial response strategy; then using equipment feedback data to calculate revision coefficients, dynamically correcting the initial strategy to form a final response strategy; controlling the hydropower station equipment according to the final strategy, and recording data to optimize the model. This method can predict disaster effects in advance, and combines multi-objective optimization with real-time equipment status for closed-loop control, effectively solving the problems of delayed emergency response, rigid strategies, and execution failure under extreme conditions in traditional emergency response, and significantly improving the safety and reliability of hydropower station earthquake emergency response. Attached Figure Description

[0043] To more clearly illustrate the technical solutions of the exemplary embodiments of this application, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of this application and should not be considered as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort. In the drawings:

[0044] Figure 1 A flowchart of an emergency control method for hydropower stations based on earthquake early warning, provided as an embodiment of this application;

[0045] Figure 2 A schematic diagram of the structure of a hydropower station emergency control system based on earthquake early warning, provided in an embodiment of this application;

[0046] The attached diagram shows the markings and corresponding component names:

[0047] 201-Data Acquisition Module, 202-Deep Learning Module, 203-Policy Generation Module, 204-Policy Correction Module, 205-Emergency Control Module, 206-Data Augmentation Module. Detailed Implementation

[0048] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the embodiments and accompanying drawings. The illustrative embodiments and descriptions of this application are only for explaining this application and are not intended to limit this application.

[0049] like Figure 1 As shown in the figure, this application provides an emergency control method for hydropower stations based on earthquake early warning, including:

[0050] S101. Acquire multi-source correlation data and preprocess it to obtain preprocessed multi-source correlation data; the multi-source correlation data includes earthquake early warning data, hydropower station operation data and historical archive data;

[0051] S102. Call the preset disaster effect prediction model to analyze the preprocessed multi-source correlation data and predict the disaster effect parameters caused by earthquakes in the future.

[0052] S103. Based on the disaster effect parameters and the current state of the hydropower station, a deep reinforcement learning algorithm is used to generate an initial response strategy for emergency control of the hydropower station.

[0053] S104. Calculate the revision coefficient using the equipment feedback data of the controllable equipment of the hydropower station, and dynamically modify the initial treatment strategy according to the revision coefficient to form the final treatment strategy;

[0054] S105. Emergency control of the controllable equipment in the hydropower station is carried out using the aforementioned final disposal strategy; the controllable equipment of the hydropower station includes at least the hydropower station gates;

[0055] S106. Record all relevant data in the current emergency control process of the hydropower station, and optimize the pre-set disaster effect prediction model and deep reinforcement learning algorithm based on all relevant data for subsequent emergency control of the hydropower station.

[0056] This application provides a method and system for emergency control of hydropower stations based on earthquake early warning. The method includes: acquiring multi-source correlated data such as earthquake early warning data, hydropower station operation data, and historical archive data, and using a disaster effect prediction model to predict disaster effect parameters caused by future earthquakes; combining the disaster effect parameters with the current state of the hydropower station, and using a deep reinforcement learning algorithm to generate an initial response strategy; then using equipment feedback data to calculate revision coefficients, dynamically correcting the initial strategy to form a final response strategy; controlling the hydropower station equipment according to the final strategy, and recording data to optimize the model. This method can predict disaster effects in advance, and combines multi-objective optimization with real-time equipment status for closed-loop control, effectively solving the problems of delayed emergency response, rigid strategies, and execution failure under extreme conditions in traditional emergency response, and significantly improving the safety and reliability of hydropower station earthquake emergency response.

[0057] Real-world earthquake scenario data is scarce. By using digital twin models to simulate various extreme conditions, the problem of insufficient training samples for reinforcement learning is solved, enabling the algorithm to have a high level of decision-making ability even before real-world application. At the same time, by using real emergency data (even minor earthquakes) to iteratively update the model, the system can continuously adapt to actual conditions such as aging hydropower station equipment and terrain changes, ensuring long-term accuracy and adaptability.

[0058] In one possible implementation, the earthquake early warning data includes earthquake intensity, epicenter location, and S-wave arrival time, which may be sourced from a seismological bureau or other earthquake prediction center; the hydropower station operation data includes hydropower station location, real-time reservoir location, flow velocity, gate opening, hoist load, motor current, voltage, generator active power, generator reactive power, dam seepage pressure and / or displacement monitoring data; the historical archive data includes the most recent historical earthquake data, historical final disposal strategies, and historical hydropower station operation data.

[0059] In one possible implementation, a pre-set disaster effect prediction model is invoked to analyze the preprocessed multi-source correlation data to predict disaster effect parameters caused by earthquakes in the future, including:

[0060] The preprocessed multi-source correlated data are combined into a vector, and the vector is used as the input of a pre-set disaster effect prediction model to predict the surge height caused by the earthquake in the future, the stress at preset key points on the hydropower station dam, and the gate impact load, so as to obtain the disaster effect parameters caused by the earthquake in the future; wherein, the disaster effect prediction model is pre-set using LSTM (Long Short-Term Memory).

[0061] In the embodiments of this application, the surge height, the stress at preset key points on the hydropower station dam, and the gate impact load can be predicted by one LSTM or by three LSTMs respectively.

[0062] In one possible implementation, the presetting method for the disaster effect prediction model includes:

[0063] The loss function is constructed as either the cross-entropy loss function or the root mean square loss function;

[0064] The disaster effect prediction model is trained and deployed with the goal of minimizing the loss function.

[0065] During the training of the disaster effect prediction model, training data can be constructed using preprocessed historical multi-source correlation data and actual disaster effect parameters. This allows for the calculation of the cross-entropy loss function or the root mean square loss function. After calculating the loss function value, the disaster effect prediction model can be trained using particle swarm optimization or gradient descent algorithms, thereby enabling the model to analyze disaster effect parameters. It is worth noting that, to achieve predictive effectiveness, the actual disaster effect parameters during training should be data from a specific point in time following the historical multi-source correlation data, with a fixed time interval between the two sets of data.

[0066] This invention, by integrating earthquake early warning data (intensity, epicenter, S-wave arrival time) and combining it with a disaster effect prediction model constructed using LSTM (Long Short-Term Memory) networks, can predict disaster effect parameters (such as surge height, dam critical point stress, and gate impact load) in advance within a very short time before the arrival of the S-wave. Traditional technologies typically trigger control only after the seismic wave arrives and sensors detect vibration, by which time the dam may have already begun to deform under stress. This invention leverages the advantages of LSTM in processing time-series data, rapidly extrapolating the dynamic evolution of the seismic wave and the reservoir water response based on multi-source correlated data. This allows the system to pre-adjust gate openings or unit loads before the destructive vibration arrives, effectively avoiding the risk of impending surges and gaining valuable time.

[0067] In one possible implementation, based on the disaster effect parameters and the current state of the hydropower station, a deep reinforcement learning algorithm is used to generate an initial emergency control strategy for the hydropower station, including:

[0068] The current reservoir water level, current inflow, current gate opening vector, and unit operating status vector of the hydropower station are obtained to obtain the current status of the hydropower station.

[0069] The disaster effect parameters and the current state of the hydropower station are constructed into a state vector, and a deep reinforcement learning algorithm is used to identify the state vector to determine the action vector; the action vector includes the opening / closing rate of the gates; it is worth noting that this refers to the opening / closing rate of all gates, a negative value indicates closure, and a positive value indicates opening.

[0070] Optionally, the action vector may also include the generator set power setpoint adjustment amount and the generator set operating mode switching command; since the generator set operating status generally does not affect safety, the generator set power setpoint adjustment amount and the generator set operating mode switching command do not need to be corrected.

[0071] The aforementioned action vector is used as the initial response strategy for emergency control of the hydropower station.

[0072] In one possible implementation, the method further includes the following steps before using the deep reinforcement learning algorithm:

[0073] Construct a digital twin simulation model of the hydropower station and obtain historical status by running the digital twin simulation model of the hydropower station;

[0074] By acquiring historical actions, historical rewards, and the next historical moment's state from historical states, a historical experience pool is constructed.

[0075] The deep reinforcement learning algorithm is optimized using the experience data in the historical experience pool, enabling the digital twin simulation model of the hydropower station to accurately select action vectors.

[0076] In one possible implementation, the historical rewards are obtained through a reward function, and the reward function is set as follows:

[0077] ;

[0078] In the formula, For the reward function, As a weighting of safety indicators, Rewards for safety performance indicators As the weight of the flood discharge index, As a reward for flood discharge, As the weight of economic indicators, Rewards for economic indicators This is a penalty item. The weight of safety indicators is greater than that of flood discharge indicators, and the weight of flood discharge indicators is greater than that of economic indicators.

[0079] The aforementioned reward functions affect the accuracy of action selection. Therefore, this application provides a detailed description of the above-mentioned rewards to enable those skilled in the art to better understand the technical solutions described in this application.

[0080] Safety performance rewards can be:

[0081] ;

[0082] In the formula, As the first safety weight, As the second safety weight, As the third safety weight, The allowable stress for the dam body design, This represents the current maximum principal stress in the dam body. This is the current highest water level. The preset check flood level, As an indicator function, under gate load Greater than the safe load threshold If the value is 1, then the value is 0; otherwise, the value is 0.

[0083] Flood discharge rewards can be:

[0084] ;

[0085] In the formula, The coefficient for calculating flood discharge rewards can be set to 0.8; This is the current discharge flow rate. For inbound flow, For the predicted surge increment, The predicted time length.

[0086] The economic indicator reward is as follows:

[0087] ;

[0088] In the formula, The coefficient for calculating economic rewards is set as a constant between (0.05, 2); This refers to the volume of water that cannot be used for power generation.

[0089] If an irreversible failure such as dam breach or equipment damage occurs during the simulation, a huge negative reward will be immediately given (e.g., =−10000) and reset the environment.

[0090] Since conducting trial-and-error training of deep reinforcement learning on a real hydropower station is impractical (extremely risky), this invention constructs a high-fidelity digital twin simulation model of a hydropower station as the training environment. This model integrates a hydrodynamic module (solving Navier-Stokes equations to simulate water flow), a structural mechanics module (solving finite element equations to simulate dam stress and strain), and an electromechanical equipment module (simulating the dynamic response of generator units and gate hoists). Virtual sensors identical to those in the physical hydropower station are deployed within the digital twin, enabling real-time feedback of data such as water level, flow velocity, stress, and power in the virtual environment. The simulation model has an API interface, capable of receiving control actions output by the agent and returning the next-moment state and reward to the agent, achieving closed-loop training of perception, decision-making, execution, and feedback.

[0091] This invention employs a deep reinforcement learning algorithm and is trained based on a reward function encompassing multiple dimensions, including safety, flood discharge, and economic considerations. It comprehensively considers complex states such as current reservoir water level, inflow, and predicted disaster effects to output the optimal action vector (e.g., gate opening rate). Traditional rule-based control is often an either-or approach (e.g., fully opening gates whenever an earthquake occurs), which may lead to unnecessary water loss or exacerbate disasters when downstream areas are already experiencing flooding. In this application, when the predicted earthquake-induced surge height is low and the reservoir water level is low, the system may choose to only slightly open the gates or adjust the generator output instead of fully opening the floodgates. This ensures that dam safety indicators (such as dam stress and gate load) remain within thresholds while also considering the economic efficiency of water resource utilization.

[0092] In one possible implementation, the revision coefficient is calculated using equipment feedback data from the controllable equipment of the hydropower station, including:

[0093] Based on the equipment feedback data from the controllable equipment of the hydropower station, obtain the real-time status adjustment factor;

[0094] Obtain the equipment health index, historical average efficiency deviation under the same operating conditions, and seasonal correction coefficient for the controllable equipment of the hydropower station, and obtain the historical experience correction factor based on the equipment health index, historical average efficiency deviation under the same operating conditions, and seasonal correction coefficient.

[0095] The revision coefficient is obtained by weighting and summing the real-time state adjustment factor and the historical experience correction factor.

[0096] In one possible implementation, the initial handling strategy is dynamically modified according to the revision coefficient to form a final handling strategy, including: multiplying the initial handling strategy by the revision coefficient to obtain the final handling strategy.

[0097] To help those skilled in the art better understand the technical solutions described in the embodiments of this application, the acquisition of the cultivation coefficient is explained in detail below.

[0098] The revision coefficient K is mainly used to revise the initial treatment strategy to obtain a more accurate final treatment strategy: ; As the initial response strategy, As a final disposal strategy;

[0099] The revision coefficient is obtained by weighted summation of the real-time state adjustment factor and the historical experience correction factor, specifically as follows: α and β are the real-time weight and historical experience weight, respectively, and satisfy α + β = 1. Usually, in emergency situations (such as the initial stage of an earthquake), α is taken as a larger value (such as 0.7~0.9) to prioritize the response to real-time physical feedback; during the stable recovery period, β is taken as a larger value. This is the environmental noise constant, used to prevent system overfitting or numerical deadlock, and is usually taken as a very small value (e.g., 0.001~0.005). For real-time state adjustment factor, Historical experience correction factor;

[0100] The real-time status adjustment factor primarily reflects the current physical health status of the hydropower station's equipment and the immediate hydrological and meteorological conditions. Taking gate opening and closing control as an example, this factor comprehensively considers mechanical load, motor current, and surge prediction difference. Therefore, the equipment feedback data of the controllable equipment in the hydropower station mainly includes real-time mechanical load, real-time motor current, and real-time surge prediction difference, and the real-time status adjustment factor can be calculated as follows:

[0101] ;

[0102] In the formula, e is the natural constant. The first preset weighting coefficient, This is the second preset weighting coefficient. The third preset weighting coefficient, This is the rated mechanical load of the hoist. For real-time mechanical load, Rated current, This is the real-time motor current. This is the real-time surge prediction difference, which is the difference between the surge height and the surge height output by the prediction model.

[0103] Historical experience correction factors are primarily used to uncover hidden patterns in historical data and correct systematic biases caused by equipment aging, seasonal water level changes, or specific geological conditions. Therefore, historical experience correction factors are obtained through equipment health index, historical average efficiency deviation under similar operating conditions, and seasonal correction coefficients, specifically:

[0104] ;

[0105] In the formula, HI is the equipment health index, and ; This is the time since the last equipment maintenance. This is the maximum maintenance cycle time for the equipment. It is the wear rate constant (derived from historical failure frequency statistics, usually taken as 0.1~0.3). This represents the deviation of the average efficiency under historical operating conditions, and ; Let be the target opening or target flow rate during the i-th historical operation. Let N be the actual opening or actual flow during the i-th historical operation, and N be the total number of historical operations. The seasonal revision factor can be set to 1.05 for summer, 1 for spring and autumn, and 0.95 for winter.

[0106] For example, suppose the system decision is: "The initial handling strategy for gate 2 is 0.1, that is, to open gate 2 at a speed of 0.1 m / s." Real-time status: Sensors report that the current hoist load has reached 90% of its rated value, and calculations show... =0.85 (speed needs to be reduced).

[0107] Historical experience: The gate has just completed its major overhaul and is in good condition (HI=1.0), but it is currently winter, and the water's viscous resistance is high. =0.95), and historical data shows that the discharge at this water level is usually slow ( =+0.02). Calculated as follows: =1.0×(1+0.02)×0.95≈0.97.

[0108] Final calculation: Assuming α and β are set to 0.8 and 0.2 respectively, If we set it to 0.001, then K = 0.8 * 0.85 + 0.2 * 0.97 + 0.001 = 0.875.

[0109] Therefore, the final handling strategy can be obtained as 0.1*0.875=0.0875, and the final instruction is to "open gate No. 2 at a speed of 0.0875m / s". This instruction not only avoids the current high load risk, but also appropriately considers the water resistance in winter and historical deviations, thus achieving precise safety control.

[0110] This invention, after generating an initial response strategy, does not execute it directly. Instead, it uses real-time feedback data (motor current, voltage, load) from controllable equipment (such as a winch-type gate opener) to calculate a revision coefficient, dynamically correcting the strategy. During an earthquake, equipment may experience increased friction due to structural deformation or a decrease in output torque due to electrical fluctuations. Simply executing an initial response strategy based on an ideal model (e.g., setting the gate to lift at 100% speed) could lead to motor overload and burnout or mechanical jamming, causing secondary disasters. This invention calculates a revision coefficient (e.g., 0.6) using a real-time status adjustment factor (reflecting current real-time load capacity) and a historical experience correction factor (reflecting equipment aging and seasonal lubrication conditions), reducing the initial opening speed to 60%. This closed-loop feedback mechanism ensures that the issued control commands (final response strategy) meet emergency requirements while remaining within the actual physical capabilities of the equipment, greatly improving the robustness and success rate of control under extreme conditions.

[0111] like Figure 2 As shown, based on the same inventive concept, this application also provides an emergency control system for hydropower stations based on earthquake early warning, including:

[0112] Data acquisition module 201 is used to acquire multi-source correlated data and preprocess it to obtain preprocessed multi-source correlated data; the multi-source correlated data includes earthquake early warning data, hydropower station operation data and historical archive data;

[0113] The deep learning module 202 is used to call a pre-set disaster effect prediction model to analyze the pre-processed multi-source correlation data and predict the disaster effect parameters caused by earthquakes in the future.

[0114] The strategy generation module 203 is used to generate an initial response strategy for emergency control of the hydropower station based on the disaster effect parameters and the current state of the hydropower station using a deep reinforcement learning algorithm.

[0115] The strategy correction module 204 is used to calculate the revision coefficient using the equipment feedback data of the controllable equipment of the hydropower station, and to dynamically correct the initial treatment strategy according to the revision coefficient to form the final treatment strategy.

[0116] Emergency control module 205 is used to perform emergency control on the controllable equipment of the hydropower station in the final disposal strategy; the controllable equipment of the hydropower station includes at least the hydropower station gates.

[0117] The data enhancement module 206 is used to record all related data in the current emergency control process of the hydropower station, and to optimize the pre-set disaster effect prediction model and deep reinforcement learning algorithm based on all related data for subsequent emergency control of the hydropower station.

[0118] This earthquake early warning-based hydropower station emergency control system can execute the above-mentioned hydropower station emergency control methods. Its principle and beneficial effects are similar, and will not be elaborated here.

[0119] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of the present invention can be implemented using various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0120] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0121] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0122] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0123] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0124] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for emergency control of hydropower stations based on earthquake early warning, characterized in that, include: Acquire multi-source correlated data and preprocess it to obtain preprocessed multi-source correlated data; The multi-source correlated data includes earthquake early warning data, hydropower station operation data, and historical archive data; The pre-set disaster effect prediction model is invoked to analyze the pre-processed multi-source correlation data and predict the disaster effect parameters caused by earthquakes in the future. Based on the disaster effect parameters and the current state of the hydropower station, a deep reinforcement learning algorithm is used to generate an initial response strategy for emergency control of the hydropower station. The revision coefficient is calculated using the equipment feedback data of the controllable equipment of the hydropower station, and the initial treatment strategy is dynamically modified according to the revision coefficient to form the final treatment strategy; The final disposal strategy is used to conduct emergency control of the controllable equipment in the hydropower station; the controllable equipment of the hydropower station includes at least the hydropower station gates; It records all relevant data during the current emergency control process of the hydropower station, and optimizes the pre-set disaster effect prediction model and deep reinforcement learning algorithm based on all relevant data for subsequent emergency control of the hydropower station.

2. The emergency control method for hydropower stations based on earthquake early warning according to claim 1, characterized in that, The earthquake early warning data includes earthquake intensity, epicenter location, and S-wave arrival time; the hydropower station operation data includes hydropower station location, real-time reservoir location, flow velocity, gate opening, hoist load, motor current, voltage, generator active power, generator reactive power, dam seepage pressure and / or displacement monitoring data; the historical archive data includes the most recent historical earthquake data, historical final disposal strategies, and historical hydropower station operation data.

3. The emergency control method for hydropower stations based on earthquake early warning according to claim 1, characterized in that, The pre-set disaster effect prediction model is invoked to analyze the preprocessed multi-source correlation data to predict disaster effect parameters caused by earthquakes in the future, including: The preprocessed multi-source correlated data are combined into a vector, and the vector is used as the input of a pre-set disaster effect prediction model to predict the surge height caused by the earthquake in the future, the stress at preset key points on the hydropower station dam, and the gate impact load, so as to obtain the disaster effect parameters caused by the earthquake in the future; wherein, the disaster effect prediction model is pre-set using LSTM.

4. The emergency control method for hydropower stations based on earthquake early warning according to claim 3, characterized in that, The pre-setting method for the disaster effect prediction model includes: The loss function is constructed as either the cross-entropy loss function or the root mean square loss function; The disaster effect prediction model is trained and deployed with the goal of minimizing the loss function.

5. The emergency control method for hydropower stations based on earthquake early warning according to claim 1, characterized in that, Based on the disaster effect parameters and the current state of the hydropower station, an initial emergency response strategy for the hydropower station is generated using a deep reinforcement learning algorithm, including: The current reservoir water level, current inflow, current gate opening vector, and unit operating status vector of the hydropower station are obtained to obtain the current status of the hydropower station. The disaster effect parameters and the current state of the hydropower station are constructed into a state vector, and a deep reinforcement learning algorithm is used to identify the state vector to determine the action vector; the action vector includes the opening / closing rate of the gate; The aforementioned action vector is used as the initial response strategy for emergency control of the hydropower station.

6. The emergency control method for hydropower stations based on earthquake early warning according to claim 5, characterized in that, Before using the deep reinforcement learning algorithm, the following are also included: Construct a digital twin simulation model of the hydropower station and obtain historical status by running the digital twin simulation model of the hydropower station; By acquiring historical actions, historical rewards, and the next historical moment's state from historical states, a historical experience pool is constructed. The deep reinforcement learning algorithm is optimized using the experience data in the historical experience pool, enabling the digital twin simulation model of the hydropower station to accurately select action vectors.

7. The emergency control method for hydropower stations based on earthquake early warning according to claim 6, characterized in that, The historical rewards are obtained through a reward function, and the reward function is set as follows: ； In the formula, For the reward function, As a weighting of safety indicators, Rewards for safety performance indicators As the weight of the flood discharge index, As a reward for flood discharge, As the weight of economic indicators, Rewards for economic indicators This is a penalty item.

8. The emergency control method for hydropower stations based on earthquake early warning according to claim 1, characterized in that, The revision coefficient is calculated using equipment feedback data from controllable devices in the hydropower station, including: Based on the equipment feedback data from the controllable equipment of the hydropower station, obtain the real-time status adjustment factor; Obtain the equipment health index, historical average efficiency deviation under the same operating conditions, and seasonal correction coefficient for the controllable equipment of the hydropower station, and obtain the historical experience correction factor based on the equipment health index, historical average efficiency deviation under the same operating conditions, and seasonal correction coefficient. The revision coefficient is obtained by weighting and summing the real-time state adjustment factor and the historical experience correction factor.

9. The emergency control method for hydropower stations based on earthquake early warning according to claim 8, characterized in that, The initial treatment strategy is dynamically modified based on the revision coefficient to form a final treatment strategy, including: multiplying the initial treatment strategy by the revision coefficient to obtain the final treatment strategy.

10. An emergency control system for hydropower stations based on earthquake early warning, characterized in that, include: The data acquisition module is used to acquire multi-source correlated data and preprocess it to obtain preprocessed multi-source correlated data. The multi-source correlated data includes earthquake early warning data, hydropower station operation data, and historical archive data; The deep learning module is used to call a pre-set disaster effect prediction model to analyze the pre-processed multi-source correlation data and predict the disaster effect parameters caused by earthquakes in the future. The strategy generation module is used to generate an initial emergency control strategy for the hydropower station based on the disaster effect parameters and the current state of the hydropower station using a deep reinforcement learning algorithm. The strategy correction module is used to calculate the revision coefficient using the equipment feedback data of the controllable equipment of the hydropower station, and to dynamically correct the initial treatment strategy according to the revision coefficient to form the final treatment strategy. An emergency control module is used to perform emergency control on the controllable equipment of the hydropower station in accordance with the final disposal strategy; the controllable equipment of the hydropower station includes at least the hydropower station gates; The data augmentation module is used to record all relevant data during the current emergency control process of the hydropower station, and to optimize the pre-set disaster effect prediction model and deep reinforcement learning algorithm based on all relevant data for subsequent emergency control of the hydropower station.

Images