Intelligent speed regulation method and device for water-turbine generator set

Through the dual speed measurement scheme and behavioral prompt diffusion strategy network generator, combined with computing tree logic verification and adaptive PID control, the problems of low speed regulation accuracy and poor adaptability of water turbine generator sets are solved, and efficient and safe multi-unit coordination control is achieved. , improving the overall efficiency of the system.

CN118934425APending Publication Date: 2024-11-12HUBEI ZHAOHENG HONGPING POWER GENERATION CO LTD
View PDF 0 Cites 10 Cited by

Patent Information

Application Number
CN202411031757.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-30
Publication Date
2024-11-12

AI Technical Summary

Technical Problem

The problems of low speed regulation accuracy, poor adaptability, insufficient safety verification and poor coordination of multiple units.

Method used

The dual speed measurement scheme is used to obtain the unit speed signal, and the generalizable policy network generator of behavior prompt diffusion generates candidate control strategies, uses the computing tree logic to verify the security and effectiveness of the policy, and uses the adaptive PID controller and coordinated control. Proportional valves and digital valves achieve speed control.

Benefits of technology

It improves the adaptability of speed measurement accuracy and control strategies, ensures system safety, optimizes the coordinated control of multiple units, improves execution accuracy and response speed, and improves the overall efficiency of the system.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118934425A_ABST
    Figure CN118934425A_ABST
Patent Text Reader

Abstract

The invention discloses an intelligent speed regulation method and device for a water-turbine generator set, and the method comprises the steps: obtaining a rotating speed signal of a set through employing a dual speed measurement scheme, and the dual speed measurement scheme comprises PT residual voltage frequency measurement and SSG fluted disc frequency measurement; carrying out fusion processing and filtering on the rotating speed signal; based on the processed rotating speed signal, generating a candidate control strategy by using a generalizable strategy network generator for behavior prompt diffusion; verifying the security and validity of the generated strategy by using calculation tree logic; performing performance evaluation on the verified strategies and selecting an optimal strategy; applying the selected optimal strategy to a self-adaptive PID (Proportion Integration Differentiation) controller; the proportional valve and the digital valve which are controlled in a coordinated mode execute control actions, the opening degree of the guide vane of the water turbine is adjusted, and speed regulation control over the water-turbine generator set is achieved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of hydroelectric power generation, and in particular to an intelligent speed regulation method and device for a hydroelectric generator set. Background Art

[0002] The hydro-turbine generator set is the core equipment of the hydroelectric power station, and its stable operation is crucial to the safety and efficiency of the power system. The traditional hydro-turbine generator set speed control system mainly adopts the PID control method. Although it has a simple structure and high reliability, its adaptability and control accuracy often cannot meet the requirements of modern power systems when facing complex and changeable operating conditions.

[0003] At present, the speed regulation technology of hydro-generator sets faces the following main problems:

[0004] Limited speed measurement accuracy: A single speed measurement solution is susceptible to interference, affecting control accuracy.

[0005] Fixed control strategy: It is difficult to adapt to different working conditions and changes in control objectives.

[0006] Insufficient safety verification: Lack of rigorous safety and effectiveness verification of the generated control strategy.

[0007] Poor coordination among multiple units: It is difficult to achieve optimal coordinated control among multiple units.

[0008] The actuator is not responsive enough: Traditional actuators cannot meet the needs of fast and precise adjustment. Summary of the invention

[0009] In view of this, the present application provides an intelligent speed control method and device for a hydro-turbine generator set, which solves the problems of low speed control accuracy, poor adaptability, insufficient safety verification and poor coordination of multiple units of the hydro-turbine generator set in the prior art.

[0010] The embodiment of the present application provides an intelligent speed regulation method for a hydro-generator set, comprising:

[0011] A dual speed measurement scheme is used to obtain the unit speed signal, and the dual speed measurement scheme includes PT residual pressure frequency measurement and SSG gear disc frequency measurement;

[0012] Performing fusion processing and filtering on the speed signal;

[0013] Based on the processed speed signal, a generalizable strategy network generator using behavioral cue diffusion is used to generate a candidate control strategy;

[0014] Use computational tree logic to verify the security and effectiveness of the generated policies;

[0015] Performing performance evaluation on the strategies that have passed the verification and selecting the optimal strategy;

[0016] Applying the selected optimal strategy to the adaptive PID controller;

[0017] The control action is executed through the coordinated control of the proportional valve and the digital valve to adjust the opening of the turbine guide vane and realize the speed control of the turbine generator set.

[0018] In the dual speed measurement scheme:

[0019] The PT residual voltage frequency measurement uses a voltage transformer to obtain a frequency signal, and the frequency measurement accuracy is 0.001Hz;

[0020] The SSG gear frequency measurement installs a gear on the shaft of the turbine generator set and uses an electromagnetic proximity switch to generate a frequency signal. The frequency measurement range is 0.2Hz-150Hz.

[0021] The fusion processing and filtering of the speed signal include:

[0022] The weighted average method is used to fuse the PT residual pressure frequency measurement and SSG gear frequency measurement signals;

[0023] The Kalman filter is used to smooth the fused signal and filter out high-frequency noise.

[0024] The generalizable policy network generator for behavior cue diffusion includes the following steps:

[0025] A behavior prompt space including discrete and continuous behavior prompts is defined, wherein the discrete behavior prompts include stable operation, fast response and energy-saving mode, and the continuous behavior prompts include target speed and expected response time;

[0026] Build a strategic network architecture, including:

[0027] An encoder network consisting of a multi-layer perceptron and a one-dimensional convolutional neural network is used to map system states and behavior cues into a latent space.

[0028] A decoder network with a variational autoencoder structure is used to generate control policy parameters from the latent space.

[0029] Implement multi-head attention mechanisms in encoders and decoders to dynamically focus on relevant state and behavior cues;

[0030] Design a noise injection scheme to inject Gaussian noise into the latent space and implement adaptive noise scaling;

[0031] Through the denoising network of the U-Net structure, the gradual denoising process from random strategy to target strategy is realized;

[0032] The policy network is trained using a multi-objective loss function consisting of reconstruction error, KL divergence, behavioral consistency loss, and adversarial loss.

[0033] Priority experience replay and importance sampling techniques are used to improve training efficiency.

[0034] Verifying the security and effectiveness of the generated policy using computational tree logic includes the following steps:

[0035] Define atomic propositions that describe the key states and control actions of the system, including speed state, power state and control actions;

[0036] Construct a temporal logic formula that represents the system characteristics, including:

[0037] Safety property formula: AG(speed≤max_speed), where speed is the current speed and max_speed is the maximum allowable speed;

[0038] Active quality formula: AF[0,T](|speed target_speed|≤ε), where target_speed is the target speed, T is the predetermined time, and ε is the allowable error;

[0039] Stability property formula: AG(AF[0,T](|speed nominal_speed|≤δ)), where nominal_speed is the rated speed and δ is the steady-state error range;

[0040] Establish a state transition diagram to represent the speed control system, and use the adjacency matrix to represent the transition relationship between states;

[0041] Implement the depth-first search algorithm, combine memoized search and bidirectional search strategies, and verify the above temporal logic formula;

[0042] Design a counterexample generation mechanism to provide a specific state sequence as a counterexample when verification fails;

[0043] Implement a lightweight online checker to continuously monitor whether key attributes satisfy the above logical formula;

[0044] Adopting incremental verification algorithm, only verifying the part of strategy changes, improving real-time verification speed;

[0045] State space representation and verification efficiency are optimized using symbolic model checking techniques including binary decision diagrams (BDDs).

[0046] Methods for evaluating the performance of validated strategies include:

[0047] Define multi-dimensional evaluation indicators, including stability indicators, dynamic performance indicators, energy efficiency indicators and robustness indicators;

[0048] Use turbine-generator model to conduct closed-loop simulation to simulate various operating conditions and disturbance scenarios;

[0049] The weighted sum method is used to synthesize various indicators, and the Pareto optimal selection algorithm is used to balance multiple objectives.

[0050] The parameter adjustment mechanism of the adaptive PID controller includes:

[0051] Based on fuzzy logic, PID parameters are adjusted online to establish a fuzzy rule base for power deviation and deviation change rate;

[0052] Use the centroid method to perform defuzzification operations and dynamically adjust the Kp, Ki, and Kd parameters;

[0053] Implement the integral separation algorithm and integral limiting strategy to prevent system oscillation caused by integral saturation.

[0054] The coordinated control strategy of proportional valves and digital valves includes:

[0055] Use proportional valve to achieve precise continuous control, driven by PWM technology with a frequency of 20kHz;

[0056] A digital valve is used to achieve fast response, and a pulse width modulation control strategy is adopted with a frequency of 200Hz;

[0057] When the deviation is less than the preset threshold, a proportional valve is used, and when the deviation is greater than the preset threshold, a digital valve is used for quick response;

[0058] Realize valve characteristic curve compensation and linearize valve response;

[0059] The valve core position feedback signal is used to build a local closed-loop control.

[0060] The method also includes implementing a multi-agent collaborative control mechanism, and the specific steps are as follows:

[0061] Establish a multi-layered control architecture, including strategic, tactical and executive layers, including:

[0062] The strategic level includes developing the overall power generation plan for the power station;

[0063] The tactical layer includes coordinating the output distribution of multiple units;

[0064] The execution layer includes executing specific speed control instructions;

[0065] Implement distributed optimization algorithms, including:

[0066] Decentralized global optimization using alternating direction multiplier method ADMM;

[0067] Design a consistency algorithm based on graph theory to ensure the coordinated operation of multiple subsystems;

[0068] Apply game theory methods, including:

[0069] Implement a Nash equilibrium solver to find optimal strategies in competitive and cooperative scenarios;

[0070] Design incentive mechanisms to promote effective collaboration among units;

[0071] Establish dynamic communication topology, including:

[0072] Implement adaptive communication strategies to dynamically adjust communication frequency and content based on system status;

[0073] Design robust information exchange protocols to cope with communication delays and packet loss;

[0074] Implement collaborative learning mechanisms, including:

[0075] Adopting a federated learning framework to achieve knowledge sharing while protecting privacy;

[0076] Design a knowledge distillation mechanism to effectively transfer the global optimal strategy to each subsystem;

[0077] Build a global state estimator, including:

[0078] Use the distributed Kalman filter algorithm to fuse multi-source information and construct the global system state;

[0079] Implement anomaly detection and fault isolation mechanisms to improve the overall robustness of the system;

[0080] Optimize load balancing and energy efficiency, including:

[0081] Implement dynamic load distribution algorithm to optimize overall energy efficiency according to the efficiency curve of each unit;

[0082] Design predictive maintenance strategies and coordinate maintenance plans for multiple units to minimize the impact on overall performance.

[0083] The embodiment of the present application also provides an intelligent speed regulating device for a hydro-generator set, comprising:

[0084] The speed measurement module adopts a dual speed measurement scheme of PT residual pressure frequency measurement and SSG gear disc frequency measurement;

[0085] A signal processing module, used for fusing and filtering the speed measurement signal;

[0086] Intelligent control module, including:

[0087] Generalizable policy network generator for behavioral cue diffusion,

[0088] Computation tree logic verification unit,

[0089] Performance Evaluation and Strategy Selection Unit;

[0090] Adaptive PID controller;

[0091] Electro-hydraulic conversion module, including proportional valves and digital valves for coordinated control;

[0092] Actuator, used to adjust the guide vane opening;

[0093] The intelligent control module generates, verifies and selects the optimal control strategy according to the system status and control objectives, and realizes efficient speed control of the hydro-generator set through the adaptive PID controller and the electro-hydraulic conversion module.

[0094] The present application also provides a computer device, the computer device comprising:

[0095] at least one processor; and,

[0096] a memory communicatively connected to the at least one processor; wherein,

[0097] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the above-mentioned intelligent speed regulation method for a hydro-generator set.

[0098] An embodiment of the present application also provides a computer-readable storage medium, which stores computer instructions, and the computer instructions are used to enable a computer to execute the above-mentioned intelligent speed regulation method for a hydro-turbine generator set.

[0099] An embodiment of the present application also provides a computer program product, including computer instructions, characterized in that when the computer instructions are executed by a processor, the steps of the above-mentioned intelligent speed regulation method for a hydro-turbine generator set are implemented.

[0100] This application has the following technical effects:

[0101] Improve speed measurement accuracy: The dual speed measurement scheme and signal fusion processing significantly improve the accuracy and reliability of speed measurement.

[0102] Enhance the adaptability of control strategies: A generalizable strategy network generator using behavior-cued diffusion can dynamically generate optimal control strategies based on different operating conditions and control objectives.

[0103] Ensure system security: Introduce a computational tree logic verification mechanism to ensure that the generated control strategy meets security and effectiveness requirements.

[0104] Optimizing the coordinated control of multiple units: Through the multi-agent collaborative control mechanism, the overall optimized operation of multiple hydro-turbine generator sets is achieved.

[0105] Improve execution accuracy and response speed: The coordinated control strategy of proportional valves and digital valves is adopted to ensure the accuracy of control and improve the rapid response capability of the system.

[0106] Achieve adaptive control: Through the adaptive PID controller, the control parameters can be dynamically adjusted according to the system status to improve the control performance.

[0107] Improve overall system efficiency: Through multi-dimensional performance evaluation and optimization algorithms, maximize power generation efficiency while ensuring system stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0108] Figure 1 is a flow chart of an intelligent speed regulation method for a hydro-generator set provided by an embodiment of the present invention;

[0109] Figure 2 is a schematic diagram of a dual speed measurement solution provided by an embodiment of the present invention;

[0110] Figure 3 It is a structural block diagram of an intelligent speed regulating device for a hydro-generator set provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0111] The embodiments of the present disclosure are described in detail below with reference to the accompanying drawings.

[0112] It should be clear that the following embodiments of the present disclosure are described by specific specific examples, and those skilled in the art can easily understand other advantages and effects of the present disclosure from the contents disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of the present disclosure, rather than all of the embodiments. The present disclosure can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present disclosure. It should be noted that the following embodiments and features in the embodiments can be combined with each other in the absence of conflict. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present disclosure.

[0113] It should be noted that various aspects of the embodiments within the scope of the appended claims are described below. It should be apparent that the aspects described herein may be embodied in a wide variety of forms, and any specific structure and / or function described herein is merely illustrative. Based on the present disclosure, it should be understood by those skilled in the art that an aspect described herein may be implemented independently of any other aspect, and two or more of these aspects may be combined in various ways. For example, any number of aspects described herein may be used to implement the device and / or practice the method. In addition, other structures and / or functionalities other than one or more of the aspects described herein may be used to implement this device and / or practice this method.

[0114] It should also be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present disclosure. The drawings only show components related to the present disclosure rather than being drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component may be changed arbitrarily, and the component layout may also be more complicated.

[0115] Additionally, in the following description, specific details are provided to facilitate a thorough understanding of the examples. However, it will be understood by those skilled in the art that the aspects described may be practiced without these specific details.

[0116] In the embodiment of the present application, the speed regulator device adopts a digital microcomputer speed regulator, which is mainly composed of the following parts:

[0117] 1. Control unit: Siemens S7-200 series PLC

[0118] 2.Electro-hydraulic conversion device: using proportional valve + digital valve mechanism

[0119] 3. Actuator: Servomotor

[0120] 4. Sensor system: including speed, position, pressure and other sensors

[0121] 5. Human-machine interface: 7-inch color LED LCD touch screen

[0122] Specifically,

[0123] This speed regulator system adopts a digital microcomputer speed regulator, which is mainly composed of the following parts:

[0124] 1. Control unit:

[0125] Adopt Siemens S7-200 series PLC

[0126] CPU model: S7-224XP, with high-speed counting and PID control functions

[0127] Analog input module: EM 235, used to collect various sensor signals

[0128] Analog output module: EM 232, used to control electro-hydraulic conversion devices

[0129] Digital input / output module: EM 223, for processing switch signals

[0130] Communication module: CP 243-1, supporting Modbus TCP protocol, used to communicate with the host computer

[0131] 2.Electro-hydraulic conversion device:

[0132] Proportional valve + digital valve mechanism

[0133] Proportional valve: Moog D633 series, rated flow 100L / min, response frequency>50Hz

[0134] Digital valve: high-speed switching valve, response time <3ms

[0135] Valve manifold: custom designed, integrated with proportional valves, digital valves, pressure sensors, etc.

[0136] 3.Executing agency:

[0137] Relay: Double-acting hydraulic cylinder, stroke 300mm, maximum working pressure 16MPa

[0138] Position feedback: built-in LVDT displacement sensor, resolution 0.01mm

[0139] 4. Sensor system:

[0140] Speed ​​sensor:

[0141] a.PT residual voltage frequency measurement: using ABB PGTR111 voltage transformer

[0142] b.SSG gear frequency measurement: Braun A5S05B50 gear speed sensor is used

[0143] Pressure sensor: WIKA A-10, range 0-20MPa, accuracy 0.5%FS

[0144] Water level sensor: Vega VEGAWELL 52, range 0-50m, accuracy 0.1% FS

[0145] Power sensor: ABB 2VAA012211A0002 power transmitter

[0146] 5. Human-machine interface:

[0147] 7-inch color LED LCD touch screen: Weintek MT8071iE

[0148] Resolution: 800x480 pixels

[0149] Support multi-touch and gesture operation

[0150] Built-in 1GHz processor and 512MB memory

[0151] 6. Power system:

[0152] Main power supply: AC 220V

[0153] Backup power supply: DC 220V

[0154] Power switching device: ABB OTM series automatic transfer switch

[0155] UPS: APC Smart-UPS1500VA LCD, providing at least 30 minutes of emergency power supply

[0156] 7. Cabinets and auxiliary equipment:

[0157] Cabinet: Rittal TS 8 series, IP55 protection grade

[0158] Air conditioner: Rittal SK 3304.540, cooling capacity 1000W

[0159] Dehumidifier: Pfannenberg PGD 20502, dehumidification capacity 1.25L / h

[0160] Lighting: LED cabinet lighting, door switch

[0161] 8. Software system:

[0162] PLC programming software: SIMATIC STEP 7-Micro / WIN SMART V2.5

[0163] HMI development software: Weintek EasyBuilder Pro

[0164] Debugging software: self-developed speed regulator parameter setting and performance analysis software

[0165] System Integration and Data Flow:

[0166] 1.PLC acts as the core control unit and interacts with various sensors and actuators through analog and digital I / O modules.

[0167] 2. The speed signal is processed by the PLC high-speed counter to achieve accurate speed measurement.

[0168] 3. The PID control algorithm is implemented inside the PLC to calculate the control output based on the given value and feedback value.

[0169] 4. The control output drives the proportional valve through the analog output module and controls the digital valve through the digital output.

[0170] 5. The relay position feedback signal is returned to the PLC through the analog input module to form a closed-loop control.

[0171] 6.HMI communicates with PLC via Modbus TCP protocol to realize data display and parameter setting.

[0172] 7. The system interacts with the power station host system through the communication module to support remote monitoring and control.

[0173] Safety design:

[0174] 1. Use a redundant speed measurement solution to ensure the reliability of the speed signal.

[0175] 2. The power supply system is equipped with UPS and automatic switching device to ensure the continuous operation of the control system.

[0176] 3. Set up an independent hardware emergency stop circuit to ensure that the guide vanes can be quickly closed when the PLC fails.

[0177] 4. Implement multiple protection logics in the software, including overspeed protection, low oil pressure protection, etc.

[0178] Scalability:

[0179] 1. Reserve enough I / O interfaces to facilitate future function expansion.

[0180] 2. The software adopts modular design, which makes it easy to add new functions or optimize control algorithms.

[0181] 3. The communication interface supports multiple protocols and can adapt to different host computer system requirements.

[0182] This supplement details the various components of the overall system architecture, including specific models, parameters and functions. The main additions include:

[0183] 1. Control unit: describes in detail the PLC model and the functions of each module.

[0184] 2.Electro-hydraulic conversion device: specifies the specific valve model and performance parameters.

[0185] 3. Actuator: Provides the specific parameters of the relay.

[0186] 4. Sensor system: lists the model and accuracy of each sensor.

[0187] 5. Human-machine interface: The specifications and performance of the touch screen are described in detail.

[0188] 6. Power supply system: Added instructions for backup power supply and UPS.

[0189] 7. Cabinets and auxiliary equipment: Added information on auxiliary equipment such as cabinets, air conditioners, and dehumidifiers.

[0190] 8. Software Systems: Lists the programming and development software used.

[0191] In addition, explanations on system integration and data flow, security design, and scalability have been added, which help to provide a more comprehensive understanding of the system's working principles and design considerations.

[0192] Embodiment 1

[0193] The present application provides an intelligent speed regulation method for a hydro-generator set, which uses the above-mentioned speed regulator to perform intelligent speed regulation, such as Figure 1 As shown, including S1-S8:

[0194] S1: Use a dual speed measurement scheme to obtain the unit speed signal;

[0195] S1.1: Use PT residual pressure to measure frequency;

[0196] S1.2: Use SSG gear disc to measure frequency;

[0197] S2: Fusion processing and filtering of the speed signal;

[0198] S2.1: Use weighted average method to fuse PT residual pressure frequency measurement and SSG gear frequency measurement signals;

[0199] S2.2: Use Kalman filter to smooth the fused signal;

[0200] S3: Generate candidate control policies using a generalizable policy network generator with behavioral cue diffusion;

[0201] S3.1: Define the behavioral cue space;

[0202] S3.2: Build a policy network architecture;

[0203] S3.3: Design a noise injection scheme;

[0204] S3.4: Implement the denoising process;

[0205] S3.5: training policy network;

[0206] S4: Use computational tree logic to verify the security and effectiveness of the generated strategy;

[0207] S4.1: Define atomic propositions;

[0208] S4.2: Construct temporal logic formulas;

[0209] S4.3: Establish a state transition diagram;

[0210] S4.4: Implement verification algorithm;

[0211] S4.5: Design a counterexample generation mechanism;

[0212] S4.6: Implement an online checker;

[0213] S4.7: Use an incremental verification algorithm;

[0214] S4.8: Use symbolic model checking techniques;

[0215] S5: Evaluate the performance of the verified strategies and select the optimal strategy;

[0216] S5.1: Define multidimensional evaluation indicators;

[0217] S5.2: Perform closed-loop simulation;

[0218] S5.3: Use a comprehensive assessment approach;

[0219] S6: applying the selected optimal strategy to the adaptive PID controller;

[0220] S6.1: Implement online adjustment of PID parameters based on fuzzy logic;

[0221] S6.2: Implement the integral separation algorithm and integral limiting strategy;

[0222] S7: Execute control actions through coordinated controlled proportional valves and digital valves;

[0223] S7.1: Achieve precise and continuous control of proportional valves;

[0224] S7.2: Realize fast response control of digital valve;

[0225] S7.3: realize valve characteristic curve compensation;

[0226] S7.4: Construct local closed-loop control;

[0227] S8: Realize multi-agent collaborative control mechanism;

[0228] S8.1: Establish a multi-layer control architecture;

[0229] S8.2: Implement distributed optimization algorithm;

[0230] S8.3: Apply game theory methods;

[0231] S8.4: Establish dynamic communication topology;

[0232] S8.5: Implement collaborative learning mechanism;

[0233] S8.6: Establish a global state estimator;

[0234] S8.7: Optimize load balancing and energy efficiency.

[0235] Embodiment 1 of the present application has the following technical effects:

[0236] Improve speed measurement accuracy: The dual speed measurement scheme and signal fusion processing significantly improve the accuracy and reliability of speed measurement.

[0237] Enhance the adaptability of control strategies: A generalizable strategy network generator using behavior-cued diffusion can dynamically generate optimal control strategies based on different operating conditions and control objectives.

[0238] Ensure system security: Introduce a computational tree logic verification mechanism to ensure that the generated control strategy meets security and effectiveness requirements.

[0239] Optimizing the coordinated control of multiple units: Through the multi-agent collaborative control mechanism, the overall optimized operation of multiple hydro-turbine generator sets is achieved.

[0240] Improve execution accuracy and response speed: The coordinated control strategy of proportional valves and digital valves is adopted to ensure the accuracy of control and improve the rapid response capability of the system.

[0241] Achieve adaptive control: Through the adaptive PID controller, the control parameters can be dynamically adjusted according to the system status to improve the control performance.

[0242] Improve overall system efficiency: Through multi-dimensional performance evaluation and optimization algorithms, maximize power generation efficiency while ensuring system stability.

[0243] Example 2

[0244] The present application provides an intelligent speed regulation method for a hydro-generator set, comprising:

[0245] S1: Use dual speed measurement scheme to obtain unit speed signal

[0246] The dual speed measurement scheme improves the accuracy and reliability of speed measurement by combining two different speed measurement technologies. This redundant design can effectively reduce the errors and instability that may be caused by a single speed measurement method.

[0247] like Figure 2As shown, the grid frequency measurement method is PT, and the unit frequency measurement method is PT residual pressure + SSG gear disk redundancy. The speed regulator has three frequency measurement loops, one for grid frequency measurement and two for machine frequency measurement. The frequency measurement range of each loop is 0.2Hz-150.000Hz. The time base used for frequency measurement is 125ns, and the frequency measurement accuracy reaches 0.001Hz.

[0248] PT residual voltage frequency measurement: The grid frequency and machine frequency sent by the unit voltage transformer PT are isolated, limited, and shaped by the two frequency measurement circuits of the speed regulator. The square wave signal is sent to the two frequency measurement input terminals of the PLC, and the frequency measurement is performed by the internal frequency measurement function instruction of the PLC.

[0249] SSG gear frequency measurement: The gear installed on the main shaft of the hydro-turbine generator set and the electromagnetic proximity switch together form a frequency signal generator. The square wave signal isolated, limited and shaped by the frequency measurement circuit inside the speed governor is sent to the PLC frequency measurement input terminal. When the main shaft of the unit rotates, it rotates with the gear, and the electromagnetic proximity switch fixed on the bracket generates a signal proportional to the unit frequency.

[0250] S1.1: Use PT residual pressure to measure frequency

[0251] PT residual voltage frequency measurement uses a potential transformer (PT) to measure the residual voltage of the generator stator winding. When the generator rotor rotates, even if the main circuit is disconnected, a weak voltage will still be induced in the stator winding. By measuring the frequency of this residual voltage, the speed of the generator can be accurately calculated.

[0252] Specific implementation steps:

[0253] 1. Install a high-precision voltage transformer and connect it to the generator stator winding;

[0254] 2. Use a high-speed data acquisition card to collect residual voltage signals, with a sampling rate of no less than 10kHz;

[0255] 3. Use the fast Fourier transform (FFT) algorithm to perform spectrum analysis on the collected signal;

[0256] 4. The frequency measurement accuracy is further improved to 0.001Hz through interpolation algorithm.

[0257] S1.2: Frequency measurement using SSG toothed disc

[0258] SSG (Speed ​​Signal Generator) gear frequency measurement is a mechanical-electrical speed measurement method. A gear is installed on the shaft of the turbine generator set, and with the electromagnetic proximity switch, a pulse signal proportional to the speed can be generated.

[0259] For example, a high-precision geared disc with no less than 60 teeth is installed on the generator shaft; a high-sensitivity electromagnetic proximity switch is installed to ensure that the detection distance is within the range of 0.1-2mm; a dedicated frequency measurement module is used to process the pulse signal output by the proximity switch; the rotational speed is calculated through pulse counting and time measurement, and the measurement range is 0.2Hz-150Hz.

[0260] Among them, S1 specifically includes:

[0261] 1. Overview of the dual speed measurement solution

[0262] Use PT residual pressure frequency measurement and SSG gear disc frequency measurement to back up each other

[0263] Both methods work simultaneously, and the more reliable signal is selected through selection logic

[0264] Implement redundant design to improve system reliability

[0265] 2. Detailed implementation of PT residual pressure frequency measurement

[0266] Use ABB PGTR111 voltage transformer, rated primary voltage 10.5kV, rated secondary voltage 100V

[0267] The PT secondary voltage is reduced to 24V through an isolation transformer to ensure PLC input safety

[0268] Use Schmitt trigger for signal shaping to convert sine wave into square wave

[0269] The high-speed counter (HSC) of the PLC is used for frequency measurement with a resolution of 0.1μs

[0270] Using sliding window algorithm, the frequency value is updated every 10ms

[0271] Software-implemented phase-locked loop (PLL) to improve frequency measurement stability

[0272] 3.Detailed implementation of SSG gear frequency measurement

[0273] Use Braun A5S05B50 gear speed sensor, measuring range 1-20,000Hz

[0274] The gear disc is made of high-strength alloy steel, has 60 teeth, and is installed at the end of the generator shaft.

[0275] The sensor output is a differential signal, which is converted to a single-ended signal through the RS422 interface

[0276] Use a programmable gain amplifier (PGA) to adjust the signal amplitude to adapt to different speeds

[0277] Digital filter is used to eliminate the noise caused by gear processing errors and shaft vibration

[0278] Another high-speed counter of the PLC is used for pulse counting, calculating the frequency every 100ms

[0279] 4. Implementation of frequency measurement accuracy and range

[0280] Frequency measurement accuracy 0.001Hz:

[0281] Use 100MHz high-precision crystal as the clock source of the PLC

[0282] Double buffering technology is used to reduce the impact of interruption delay on measurement

[0283] Implement temperature compensation algorithm to eliminate the influence of temperature drift

[0284] Frequency measurement range 0.2Hz-150Hz:

[0285] Low frequency band (0.2Hz-10Hz): using period measurement method

[0286] Mid-frequency band (10Hz-100Hz): frequency measurement method

[0287] High frequency band (100Hz-150Hz): multi-cycle averaging method is used

[0288] Adaptive measurement algorithm: automatically selects the best measurement method based on the current frequency

[0289] 5. Signal processing and filtering

[0290] A Butterworth low-pass filter is used with a cutoff frequency of 1 / 10 of the measurement frequency.

[0291] Implement adaptive median filtering to effectively remove sudden interference

[0292] Use Kalman filter algorithm to optimize speed estimation accuracy

[0293] 6. Selection and switching of speed measurement signal

[0294] Design a speed measurement signal credibility assessment algorithm, taking into account factors such as signal stability and rate of change

[0295] Implement a disturbance-free switching strategy to ensure smooth switching between the two speed measurement methods

[0296] An interpolation algorithm is used during the switching process to avoid speed signal jumps

[0297] 7. Handling of abnormal situations

[0298] PT signal loss: automatically switch to SSG gear frequency measurement and trigger warning

[0299] Abnormal gear signal: abnormal pulse intervals of three consecutive cycles are detected, and the PT frequency measurement is switched.

[0300] Double signal failure: Start the preset safety strategy, such as maintaining the last valid speed value or executing a safe stop

[0301] 8. Performance indicators

[0302] Static accuracy: ±0.01% rated speed

[0303] Dynamic response: measurement delay <10ms for step changes

[0304] Anti-interference ability: Under the vibration condition of 5% of the rated speed, the measurement error is <0.05%

[0305] S1 can ensure that the speed regulator obtains accurate and stable speed signals, providing reliable input for subsequent control algorithms.

[0306] In addition, in the embodiment of the present invention, the speed regulator can also perform acceleration calculation, for example:

[0307] The speed value is sampled every 10ms by using the PLC high-speed counter function. The instantaneous acceleration is calculated by dividing the difference between two adjacent speed values ​​by the time interval. The acceleration data is smoothed by a digital filtering algorithm to reduce the impact of noise.

[0308] Specifically,

[0309] 1. High-speed data acquisition

[0310] Utilize the high-speed counter (HSC) function of Siemens S7-200 series PLC.

[0311] Configure the HSC to periodic capture mode to trigger an interrupt every 10ms.

[0312] The current speed value can be read in the interrupt service program with an accuracy of up to 0.1rpm.

[0313] Use double buffering technology to store speed data to ensure data continuity and integrity. Implement data timestamp function to accurately record the time of each speed sampling point.

[0314] 2. Instantaneous acceleration calculation

[0315] The instantaneous acceleration is calculated using the central difference method:

[0316] a(t)=[v(t+Δt)v(t-Δt)] / (2Δt)

[0317] Where a(t) is the acceleration at time t, v(t) is the velocity at time t, and Δt is the sampling interval (10ms)

[0318] Use 32-bit floating point numbers for calculations to ensure accuracy.

[0319] Implement dynamic time interval adjustment to adapt to possible sampling delays.

[0320] Acceleration Unit Conversion: Convert rpm / s to rad / s 2 , which is convenient for subsequent control algorithms.

[0321] 3. Digital filtering algorithm

[0322] Implement a multi-stage filtering strategy to gradually reduce the impact of noise:

[0323] a. Median filter: Use 5-point median filter to remove sudden interference.

[0324] b. Low-pass filtering: Use a second-order Butterworth low-pass filter with a cut-off frequency of 25Hz.

[0325] c. Kalman filter: Implement the adaptive Kalman filter algorithm to dynamically estimate the optimal acceleration value. The specific implementation of the low-pass filter includes the following formula:

[0326] y(n)=a0*x(n)+a1*x(n-1)+a2*x(n-2)b1*y(n-1)b2*y(n-2)

[0327] Where x is the input, y is the output, and a and b are the filter coefficients.

[0328] The key parameters of Kalman filter are:

[0329] State transfer matrix: based on uniform acceleration model;

[0330] Measurement noise covariance R: dynamically adjusted according to the accuracy of speed measurement;

[0331] Process noise covariance Q: determined by optimizing experimental data;

[0332] 4. Acceleration estimation optimization

[0333] Implement acceleration prediction algorithms based on historical data and current trends;

[0334] The weighted least square method is used to fit the velocity curve to improve the smoothness of acceleration estimation;

[0335] Implement outlier detection and elimination mechanism to prevent acceleration estimation deviation caused by sudden interference; 5. Adaptive sampling rate adjustment

[0336] Dynamically adjust the sampling frequency according to the speed change rate:

[0337] During steady-state operation: reduce the sampling rate to 50Hz to reduce the computational load;

[0338] During transients: Increase the sampling rate to 200 Hz to capture rapid changes;

[0339] Implement a smooth sampling rate switching algorithm to avoid jumps in acceleration calculation;

[0340] 6. Data Storage and Analysis

[0341] Use the data logging function of the PLC to store the speed and acceleration data of the last hour;

[0342] Implement a circular buffer to ensure that long-term operation does not exhaust memory;

[0343] Provides data export interface and supports CSV format for offline analysis;

[0344] 7. Performance indicators

[0345] Acceleration calculation delay: <5ms;

[0346] Acceleration resolution: 0.01rad / s 2 ;

[0347] Noise level: <0.1rad / s 2 (under steady-state conditions);

[0348] Dynamic response: can accurately capture 50rad / s 2 The acceleration changes.

[0349] 8. Exception handling mechanism

[0350] Design an acceleration limiting algorithm to prevent acceleration calculation anomalies caused by speed measurement errors; implement acceleration jump detection and trigger a warning when an abnormal jump is detected;

[0351] Backup strategy when acceleration signal is lost: use the last valid acceleration value and activate the corresponding protection logic.

[0352] S2: Fusion processing and filtering of the speed signal

[0353] Signal fusion and filtering are designed to comprehensively utilize the advantages of the two speed measurement methods, while filtering out noise in the measurement to provide a more accurate and stable speed signal.

[0354] S2.1: Use weighted average method to fuse PT residual pressure frequency measurement and SSG gear frequency measurement signals

[0355] The weighted average method dynamically adjusts the weights of the two speed measurement methods in the final result according to their reliability under different working conditions.

[0356] S2.1 includes:

[0357] Define weight factors w1 and w2, with initial values ​​of 0.5;

[0358] The fused speed signal S = w1 S_PT + w2 S_SSG, where S_PT is the PT residual pressure frequency measurement signal, and S_SSG is the SSG gear frequency measurement signal;

[0359] Monitor the stability and consistency of the two speed measurement methods in real time; dynamically adjust the weight factor according to the monitoring results. For example, when an abnormality is detected in the PT residual pressure frequency measurement signal, reduce w1 and increase w2.

[0360] S2.2: Use Kalman filter to smooth the fused signal

[0361] The Kalman filter is a recursive optimal estimator that can effectively remove high-frequency noise and provide a smooth speed signal.

[0362] S2.2 includes:

[0363] Establish the system state equation and observation equation:

[0364] x(k)=Ax(k-1)+w(k-1)

[0365] z(k)=Hx(k)+v(k)

[0366] Where x(k) is the true speed at time k, z(k) is the observed value, w(k) and v(k) are process noise and observation noise respectively;

[0367] Initialize the state estimate and error covariance matrix;

[0368] Prediction step: Calculate the prior state estimate and the prior error covariance;

[0369] Update step: calculate the Kalman gain, update the posterior state estimate and the posterior error covariance;

[0370] The prediction and update steps are iteratively performed to output a filtered speed signal.

[0371] S3: Generating candidate control policies using a generalizable policy network generator with behavioral cue diffusion

[0372] The generalizable policy network generator with behavior-cued diffusion is a deep learning-based intelligent control strategy generation method that can generate adaptive control strategies based on the current system state and the desired control behavior.

[0373] S3.1: Define the behavioral cue space

[0374] The behavioral cue space contains two types of cues, discrete and continuous, which are used to guide the strategy generation process.

[0375] The specific steps include:

[0376] Define a discrete behavior prompt set D = {stable operation, fast response, energy-saving mode};

[0377] Define the continuous behavior prompt set C = {target speed, expected response time};

[0378] Encode discrete cues into one-hot vectors and normalize continuous cues to the interval [0,1];

[0379] Combine discrete and continuous cues to form a complete behavioral cue vector B.

[0380] S3.2: Building a policy network architecture

[0381] The policy network adopts an encoder-decoder structure combined with an attention mechanism to achieve the mapping from system status and behavior cues to control strategies.

[0382] The specific implementation steps include:

[0383] 1. Encoder Network:

[0384] Use multi-layer perceptron (MLP) to process system status and behavior prompts; use one-dimensional convolutional neural network (1DCNN) to extract time series features; implement multi-head attention mechanism to focus on important status and prompt information;

[0385] 2. Decoder network:

[0386] A variational autoencoder (VAE) structure is adopted to enhance the diversity of generation strategies. A multi-layer MLP is used to generate control strategy parameters from the latent space. A multi-head attention mechanism is implemented to dynamically focus on relevant latent features.

[0387] Specifically, 1. Encoder network:

[0388] a) Multi-layer Perceptron (MLP) processes system status and behavior cues:

[0389] Input layer: receives system state vector (such as speed, power, head, etc.) and behavior prompt vector

[0390] Hidden layer: Use 3 fully connected layers, each with 256, 128, and 64 neurons respectively

[0391] Activation function: Use ReLU activation function to improve the nonlinear expression ability of the network

[0392] Batch Normalization: Add a batch normalization layer after each hidden layer to speed up training and improve generalization ability

[0393] b) One-dimensional convolutional neural network (1D CNN) extracts temporal features:

[0394] Input: The system states of the last N time steps are combined into a time series

[0395] Convolutional layers: Use 3 1D convolutional layers with kernel sizes of 3, 5, and 7, and 64 filters per layer. Pooling layers: Use max pooling after each convolutional layer with a pooling size of 2.

[0396] Activation function: Use LeakyReLU, parameter α = 0.1

[0397] c) Multi-head attention mechanism:

[0398] Concatenate the outputs of MLP and CNN as the input of the attention mechanism

[0399] Use 8 attention heads, each with a dimension of 32

[0400] Attention calculation: Q = WqX, K = WkX, V = WvX

[0401] Attention(Q,K,V)=softmax(QK^T / √dk)V

[0402] Output: Concatenate the outputs of the 8 attention heads and reduce the dimension through a linear layer

[0403] 2. Decoder network:

[0404] a) Variational Autoencoder (VAE) structure:

[0405] The encoder outputs μ and logσ, which represent the mean and variance of the latent space.

[0406] Reparameterization technique: z = μ + exp(logσ)*ε, where ε ~ N(0,I)

[0407] KL divergence loss: DKL = 0.5*Σ(exp(logσ)+μ^2 1logσ)

[0408] b) Multi-layer MLP generates control strategy parameters:

[0409] Input layer: receives the sampled potential vector z;

[0410] Hidden layer: 4 fully connected layers, each with 64, 128, 256, and 128 neurons respectively;

[0411] Output layer: Generate control strategy parameters, such as PID parameters, valve opening, etc.;

[0412] Activation function: The hidden layer uses the ELU activation function, and the output layer uses tanh to ensure that the output is in the range of [-1, 1].

[0413] c) Multi-head attention mechanism:

[0414] Implement multi-head attention on the middle layer of MLP (128-dimensional output);

[0415] Use 4 attention heads, each with a dimension of 32;

[0416] Self-attention computation: allows the network to focus on important features in the latent space;

[0417] Residual connection: Add the attention output to the original input to prevent gradient disappearance;

[0418] d) Additional design considerations:

[0419] Introducing Skip Connection: directly connecting the intermediate layer features of the encoder to the corresponding layer of the decoder;

[0420] Use Dropout: Add a Dropout layer between the fully connected layers with a ratio of 0.1 to improve the generalization ability of the model;

[0421] Weight initialization: Use the He initialization method, which is suitable for ReLU and its variant activation functions;

[0422] e) Output processing:

[0423] Design special output heads for different types of control parameters;

[0424] Use an appropriate activation function, such as sigmoid for parameters in the range [0,1] and tanh for parameters in the range [-1,1];

[0425] Add a scaling layer to map the network output to the valid range of the actual control parameters.

[0426] Through this carefully designed encoder-decoder structure, the network can effectively extract key information from system states and behavior cues and generate diverse, high-quality control strategy parameters. The use of a multi-head attention mechanism enables the network to dynamically focus on important states and features, improving the pertinence and effectiveness of strategy generation.

[0427] S3.3: Design a noise injection scheme

[0428] Inject noise into the latent space to increase the exploratory and robustness of the generation strategy. Inject Gaussian noise into the latent space of VAE: z = μ + σε, where ε ~ N(0, I); achieve adaptive noise scale: σ = f(s, B), where f is the neural network, s is the system state, and B is the behavior prompt; gradually reduce the noise level during training to achieve a smooth transition from exploration to utilization.

[0429] S3.4: Implementing the denoising process

[0430] A denoising network with a U-Net structure is used to gradually refine the random strategy to the target strategy.

[0431] Specifically include:

[0432] Construct a denoising network with a U-Net structure, including downsampling, upsampling and skip connections; define the number of denoising steps T, and initialize T denoising networks Dt, t = 1, 2, ..., T;

[0433] For each denoising step t:

[0434] Input: the strategy parameter θt-1 of the previous step and the current noise level σt;

[0435] Output: Update policy parameters θt = Dt(θt-1,σt);

[0436] The final output θt is used as the generated control strategy parameter.

[0437] S3.5: Training the policy network

[0438] Multi-objective loss functions and efficient training techniques are used to improve the performance and training efficiency of policy networks.

[0439] Specific implementation steps:

[0440] 1. Define the multi-objective loss function:

[0441] L=λ1L_recon+λ2L_KL+λ3L_behavior+λ4L_adv

[0442] Where L_recon is the reconstruction error, L_KL is the KL divergence, L_behavior is the behavior consistency loss, and L_adv is the adversarial loss;

[0443] 2. Implement priority experience playback:

[0444] Maintain an experience pool to store (s, B, θ, r) quadruplets; calculate the TD error of each sample as a priority indicator; sample samples according to priority and update network parameters;

[0445] 3. Use Importance Sampling:

[0446] Calculate the importance weight w = (1 / N 1 / P(i))^β, where N is the batch size, P(i) is the probability that sample i is selected, and β is the annealing parameter of importance sampling; use the weight w to weight the loss function;

[0447] 4. Use gradient clipping technology to prevent gradient explosion;

[0448] 5. Use Adam optimizer to update parameters, and the learning rate adopts cosine annealing strategy.

[0449] Through the above steps, the generalizable policy network generator with behavior-cue diffusion can generate adaptable candidate control strategies according to the current system state and the desired control behavior.

[0450] S4: Use computational tree logic to verify the security and effectiveness of the generated strategy

[0451] Computational Tree Logic (CTL) verification is a formal method used to rigorously verify the safety and effectiveness of control strategies. This step ensures that the generated control strategy meets the predetermined safety constraints and performance requirements.

[0452] S4.1: Define atomic propositions

[0453] First, we define atomic propositions that describe the key states and control actions of the system:

[0454] 1. Speed ​​status: speed_normal, speed_high, speed_low

[0455] 2. Power status: power_normal, power_high, power_low

[0456] 3. Control actions: action_increase, action_decrease, action_maintain

[0457] S4.2: Constructing Temporal Logic Formulas

[0458] Based on the defined atomic propositions, construct temporal logic formulas that represent system characteristics:

[0459] 1. Safety property formula: AG (speed ≤ max_speed)

[0460] Meaning: At any time, the speed does not exceed the maximum allowable speed.

[0461] 2. Active quality formula: AF[0,T](|speed target_speed|≤ε)

[0462] Meaning: Within the predetermined time T, the system can eventually reach the ε neighborhood of the target speed.

[0463] 3. Stability property formula: AG(AF[0,T](|speed nominal_speed|≤δ))

[0464] Meaning: The system can always return to the δ neighborhood of the rated speed within time T.

[0465] S4.3: Create a state transition diagram

[0466] 1. Define the system state set S, including key parameters such as speed and power;

[0467] 2. Define the action set A, including control actions such as increasing, decreasing, and maintaining the guide vane opening;

[0468] 3. Construct a state transition function T: S × A → S to describe the change of system state under a given action;

[0469] 4. Use the adjacency matrix M to represent the transition relationship between states, where M[i][j]=1 means that state i can be transformed to state j.

[0470] S4.4: Implement verification algorithm

[0471] 1. Implement the Depth First Search (DFS) algorithm:

[0472] Starting from the initial state, all possible state transitions are recursively explored; during the search, each state is checked to see if it satisfies the CTL formula.

[0473] 2. Combined with memorization search strategy:

[0474] Use a hash table to store verified sub-formula results; avoid repeated calculations and improve verification efficiency.

[0475] 3. Implement a bidirectional search strategy:

[0476] Start the search from both the initial state and the goal state;

[0477] When two search directions meet, a complete state path is constructed.

[0478] For example,

[0479] 1. Depth-first search (DFS) algorithm implementation:

[0480] ```Python

[0481] def dfs_verify(state,formula):

[0482] if state in visited:

[0483] return True

[0484] visited.add(state)

[0485] if state.speed>MAX_SPEED:

[0486] return False

[0487] for action in get_possible_actions(state):

[0488] next_state = apply_action(state,action)

[0489] if not dfs_verify(next_state,formula):

[0490] return False

[0491] return True

[0492] MAX_SPEED=3000#rpm

[0493] visited=set()

[0494] initial_state=State(speed=2800,power=0.8,valve_position=0.6)result=dfs_verify(initial_state,"AG(speed<=max_speed)")

[0495] print("验证结果:","通过"if result else"失败")

[0496] ```

[0497] 2. Memory-based search strategy:

[0498] ```python

[0499] def memo_dfs_verify(state,formula,memo = None):if memo is None:

[0500] memo={}

[0501] state_key = hash(state)

[0502] if state_key in memo:

[0503] return memo[state_key]

[0504] if state.speed > MAX_SPEED:

[0505] memo[state_key] = False

[0506] return False for action in get_possible_actions(state):

[0507] next_state = apply_action(state, action)

[0508] if not memo_dfs_verify(next_state, formula, memo):

[0509] memo[state_key] = False

[0510] return False memo[state_key] = True

[0511] return True

[0512] result = memo_dfs_verify(initial_state, "AG(speed <= max_speed)")```

[0513] 3. Bidirectional search strategy implementation:

[0514] ```python

[0515] def bidirectional_search(initial_state, goal_predicate): forward_queue = deque([initial_state])

[0516] backward_queue = deque(get_goal_states(goal_predicate))

[0517] forward_visited = set()

[0518] backward_visited = set()

[0519] while forward_queue and backward_queue:

[0520] # Forward search

[0521] for _ in range(len(forward_queue)):

[0522] state = forward_queue.popleft()

[0523] if state in backward_visited:

[0524] return True # Path found

[0525] forward_visited.add(state)

[0526] for next_state in get_next_states(state):

[0527] if next_state not in forward_visited:

[0528] forward_queue.append(next_state)

[0529] # Backward search

[0530] for _ in range(len(backward_queue)):

[0531] state = backward_queue.popleft()

[0532] if state in forward_visited:

[0533] return True # Path found

[0534] backward_visited.add(state)

[0535] for prev_state in get_prev_states(state):

[0536] if prev_state not in backward_visited:

[0537] backward_queue.append(prev_state)

[0538] return False#path not found

[0539] def get_goal_states(goal_predicate):

[0540] #Return the state set that satisfies the target predicate

[0541] return[State(speed=MAX_SPEED,power=p,valve_position=v)

[0542] for p in np.linspace(0,1,10)

[0543] for v in np.linspace(0,1,10)]

[0544] result=bidirectional_search(initial_state,lambda s:s.speed<=MAX_SPEED)

[0545] ```

[0546] Specific examples:

[0547] Assuming there is a simplified state space of a hydro-generator set, including speed, power and valve position, the embodiment of the present application is to verify whether the speed is always kept below 3000 rpm in all possible state transitions.

[0548] 1. Initial state: speed = 2800rpm, power = 0.8, valve_position = 0.6

[0549] 2. Possible actions: increase valve opening, decrease valve opening, keep unchanged

[0550] 3. State transition function (simplified version):

[0551] If the valve opening is increased: speed+=100, power+=0.05, valve_position+=0.1

[0552] If the valve opening is reduced: speed-=100, power-=0.05, valve_position-=0.1

[0553] If unchanged: Status unchanged

[0554] Verification process:

[0555] 1. Starting from the initial state, the DFS algorithm explores all possible state transitions.

[0556] 2. For each accessed state, check if the speed exceeds 3000rpm.

[0557] 3. If any state is found to violate the constraints, verification failure is returned immediately.

[0558] 4. If all reachable states meet the constraints, return verification success.

[0559] ```Python

[0560] def verify_speed_constraint(state,visited=None):

[0561] if visited is None:

[0562] visited=set()

[0563] state_key=(state.speed,state.power,state.valve_position)

[0564] if state_key in visited:

[0565] return True

[0566] visited.add(state_key)

[0567] if state.speed>3000:

[0568] print(f"Constraint violation status:{state}")

[0569] return False

[0570] actions=["increase","decrease","maintain"]

[0571] for action in actions:

[0572] next_state=apply_action(state,action)

[0573] if not verify_speed_constraint(next_state,visited):

[0574] return False

[0575] return True

[0576] class State:

[0577] def__init__(self,speed,power,valve_position):

[0578] self.speed=speed

[0579] self.power=power

[0580] self.valve_position=valve_position

[0581] def__str__(self):

[0582] return f"Speed:{self.speed},Power:{self.power:.2f},Valve:{self.valve_position:.2f}"

[0583] def apply_action(state,action):

[0584] new_state=State(state.speed,state.power,state.valve_position)

[0585] if action=="increase":

[0586] new_state.speed+=100

[0587] new_state.power=min(1.0,new_state.power+0.05)

[0588] new_state.valve_position=min(1.0,new_state.valve_position+0.1)

[0589] elif action=="decrease":

[0590] new_state.speed -= 100

[0591] new_state.power=max(0.0,new_state.power 0.05)

[0592] new_state.valve_position=max(0.0,new_state.valve_position 0.1)

[0593] return new_state

[0594] initial_state=State(2800,0.8,0.6)

[0595] result=verify_speed_constraint(initial_state)

[0596] print("Verification result:","Passed"if result else"Failed")

[0597] ```

[0598] S4.5: Design a counterexample generation mechanism

[0599] When verification fails, the search path is backtracked to construct the state sequence that leads to the violation of the property; the counterexample sequence is converted into a human-readable form, including key state parameters and control actions; and a counterexample analysis tool is provided to help developers understand and correct problems in the strategy.

[0600] S4.6: Implementing an online checker

[0601] A lightweight CTL formula evaluator is designed to support real-time checking of key safety properties. The online checker is updated with the current status in each control cycle of the control system. When a potential safety violation is detected, an alarm mechanism is triggered and an emergency control strategy is initiated.

[0602] S4.7: Use incremental verification algorithm

[0603] Record the results of the last verification, including the verified state set and satisfied sub-formulas; for newly generated policies, only verify the changed parts: identify the state set affected by the policy change; only re-verify these states and their successor states. Reuse the verification results of the unaffected parts to significantly improve the real-time verification speed.

[0604] S4.8: Use symbolic model checking techniques

[0605] 1. Use binary decision diagram (BDD) to represent system states and transition relationships:

[0606] Encode state variables as binary variables; use BDD packages such as the CUDD library to build and manipulate decision diagrams.

[0607] 2. Implement the symbolic reachability analysis algorithm:

[0608] Use forward Image and backward PreImage operations to calculate the set of reachable states; use BDD operations to efficiently handle large state spaces.

[0609] 3. Optimize the order of BDD variables, reduce the size of BDD, and improve verification efficiency.

[0610] Through the above steps, the system can strictly verify the security and effectiveness of the generated strategy to ensure that it meets the predetermined timing logic properties.

[0611] S5: Evaluate the performance of the verified strategies and select the optimal strategy

[0612] Performance evaluation and strategy selection are key steps to ensure that the selected control strategy is not only safe and effective, but also achieves optimal performance.

[0613] S5.1: Define multidimensional evaluation metrics

[0614] 1. Stability index:

[0615] Steady-state error: |steady_speed target_speed|

[0616] Overshoot: (max_speed target_speed) / target_speed 100%

[0617] Adjustment time: The time required to reach and maintain the target speed within ±2%.

[0618] 2. Dynamic performance indicators:

[0619] Rise Time: The time required to reach 90% of the target value from the initial state.

[0620] Peak time: The time when the first peak is reached.

[0621] Attenuation ratio: the ratio between adjacent peaks.

[0622] 3. Energy efficiency indicators:

[0623] Power generation efficiency: output electrical energy / input mechanical energy.

[0624] Guide vane opening change rate: Σ|Δθ| / T, where θ is the guide vane opening and T is the total time.

[0625] 4. Robustness indicators:

[0626] Disturbance suppression capability: the maximum deviation under a given amplitude disturbance.

[0627] Parameter sensitivity: How sensitive the control performance is to changes in system parameters.

[0628] S5.2: Perform closed-loop simulation

[0629] 1. Construct a high-fidelity mathematical model of the turbine-generator set:

[0630] Including the hydraulic characteristics of the turbine, the electromagnetic characteristics of the generator, the dynamic characteristics of the speed control system, etc.

[0631] Consider nonlinear factors such as dead zone, saturation, hysteresis, etc.

[0632] 2. Design typical working conditions and disturbance scenarios:

[0633] Normal startup and shutdown process.

[0634] Load mutation: ±30% step change of rated load.

[0635] Head change: ±10% sine wave disturbance.

[0636] Island operation: frequency regulation after grid disconnection.

[0637] 3. Run Monte Carlo simulation:

[0638] The system parameters are randomly perturbed to simulate the uncertainty in actual operation.

[0639] Each strategy was run multiple times under each condition and statistical data were collected.

[0640] For example, closed-loop simulation is a key step in evaluating the performance of a hydro-generator control strategy. This process simulates the dynamic behavior of the actual system, allowing the embodiments of the present application to test and optimize the control strategy under different conditions.

[0641] The following are the main steps and considerations for closed-loop simulation:

[0642] 1. Construct the mathematical model of turbine-generator set:

[0643] Including the hydraulic characteristic equation of the turbine

[0644] Generator rotor motion equation

[0645] Dynamic characteristic equation of speed control system

[0646] Account for nonlinearities such as dead zone, saturation, and hysteresis effects

[0647] 2. Design typical working conditions and disturbance scenarios:

[0648] Normal start process: simulates the process from standstill to rated speed

[0649] Load mutation: simulates a sudden increase or decrease in grid load, such as a ±30% step change

[0650] Head change: simulate the head fluctuation caused by reservoir water level change, such as ±10% sine wave disturbance

[0651] Island operation: simulating the frequency regulation process after grid disconnection

[0652] Grid-connected operation: Simulates power regulation when operating in parallel with the grid

[0653] 3. Implement control algorithms:

[0654] Apply the generated control strategy (such as optimized PID parameters) to the simulation model

[0655] Implement adaptive control logic to dynamically adjust control parameters according to system status

[0656] 4. Run the time series simulation:

[0657] Choose an appropriate simulation duration, usually a few minutes to a few hours

[0658] Solve differential equations using suitable numerical integration methods (such as the Runge-Kutta method)

[0659] Record the time series of key state variables, such as speed, power, guide vane opening, etc.

[0660] 5. Introducing random factors:

[0661] Randomly perturb system parameters to simulate the uncertainty in actual operation

[0662] Use the Monte Carlo method to perform multiple simulations and obtain statistical results

[0663] 6. Performance index calculation:

[0664] Steady-state error: Steady-state deviation under rated conditions

[0665] Settling time: The time required for the system to reach and remain within ±2% of the target value.

[0666] Overshoot: The ratio of the maximum deviation to the final steady-state value

[0667] Integrated absolute error (IAE): ∫|e(t)|dt, evaluates the overall control effect

[0668] Energy efficiency: the ratio of output electrical energy to input mechanical energy

[0669] 7. Extreme working condition test:

[0670] Simulate extreme hydrological conditions such as floods or droughts

[0671] Test system performance under maximum and minimum load

[0672] Evaluate the robustness of control strategies in the presence of equipment aging or partial failure

[0673] 8.Multi-unit coordination simulation:

[0674] Simulate the interaction between multiple units

[0675] Testing the effectiveness of the load distribution algorithm

[0676] Evaluate the overall performance of the system under certain unit failure conditions

[0677] 9. Long-term running simulation:

[0678] Run simulations for longer periods of time (e.g., days or weeks) to evaluate the long-term stability of control strategies

[0679] Simulate daily load variation patterns and seasonal hydrological changes

[0680] 10. Result analysis and visualization:

[0681] Draw a time series graph of key variables

[0682] Calculate and compare performance metrics under different strategies

[0683] Generate Pareto front and display multi-objective optimization results

[0684] Through this comprehensive closed-loop simulation process, the embodiments of the present application can comprehensively evaluate the performance of the control strategy, identify potential problems, and make necessary optimizations and adjustments before actual application. This not only improves the reliability and efficiency of the system, but also reduces the risk of implementing new control strategies.

[0685] S5.3: Use a comprehensive assessment approach

[0686] 1. Calculate the weighted sum:

[0687] Score=Σ(wi Normi(Indicatori))

[0688] Where wi is the weight, Normi ​​is the normalization function, and Indicatori is the i-th indicator value

[0689] 2. Use the Pareto optimal selection algorithm:

[0690] Identify the set of non-dominated solutions, i.e., solutions that cannot be improved on any one metric without compromising other metrics.

[0691] In the non-dominated solution set, the final strategy is selected according to the decision maker's preference

[0692] 3. Implement interactive decision support:

[0693] Visualize multi-dimensional evaluation results, such as radar charts and scatter matrix charts

[0694] Allow decision makers to dynamically adjust indicator weights and observe the impact on the optimal strategy selection

[0695] Through this comprehensive evaluation and selection process, the system is able to select the control strategy with the best performance while meeting safety and effectiveness requirements.

[0696] S6: Apply the selected optimal strategy to the adaptive PID controller

[0697] The adaptive PID controller can dynamically adjust its parameters according to the system state and control objectives, thus achieving more precise and stable control. Applying the optimal strategy to the adaptive PID controller can give full play to the advantages of intelligent strategy generation and classical control methods.

[0698] S6.1: Implement online adjustment of PID parameters based on fuzzy logic

[0699] 1. Define input variables:

[0700] e: Speed ​​deviation (target speed actual speed)

[0701] ec: speed deviation change rate

[0702] 2. Define output variables:

[0703] ΔKp: Proportional gain adjustment

[0704] ΔKi: Integral gain adjustment

[0705] ΔKd: Differential gain adjustment

[0706] 3. Design fuzzy rule base:

[0707] For example: IF(e is PB)AND(ec is NB)THEN(ΔKp is PB)(ΔKiis NB)(ΔKd is PS)

[0708] A total of 49 rules are designed to cover various combinations of e and ec

[0709] 4. Fuzzy reasoning process:

[0710] Using Mamdani reasoning method

[0711] Using the min-max composition operator

[0712] 5. Defuzzification:

[0713] Use the centroid method to calculate the exact values ​​of ΔKp, ΔKi, and ΔKd

[0714] 6. Update PID parameters:

[0715] Kp=Kp0+ΔKp

[0716] Ki=Ki0+ΔKi

[0717] Kd=Kd0+ΔKd

[0718] Among them, Kp0, Ki0, Kd0 are the initial parameter values

[0719] S6.2: Implementing the integral separation algorithm and integral limiting strategy

[0720] 1. Integral separation algorithm:

[0721] When |e|>e_threshold,

[0722] u(t)=Kpe(t)+Kdde(t) / dt

[0723] When |e|≤e_threshold,

[0724] u(t)=Kp e(t)+Ki∫e(t)dt+Kd de(t) / dt

[0725] 2. Points limit strategy:

[0726] If(∫e(t)dt>I_max)Then∫e(t)dt=I_max

[0727] If(∫e(t)dt<-I_max)Then∫e(t)dt=-I_max

[0728] 3. Dynamically adjust the integral limit value:

[0729] I_max=f(|e|), where f is a monotonically decreasing function

[0730] When the deviation is large, reduce the integral action to prevent excessive overshoot

[0731] 4. Realize the anti-saturation mechanism:

[0732] When the controller output reaches saturation, the integral action is stopped to prevent large oscillations caused by integral saturation.

[0733] Through these technologies, the adaptive PID controller can dynamically adjust its parameters according to the system state, effectively prevent integral saturation, and improve control performance and robustness.

[0734] In addition, in another embodiment, the PID controller can also perform output adjustment and speed adjustment, wherein the output adjustment includes steps 1-4, and the speed adjustment includes steps 5-8:

[0735] Step 1: Power Measurement

[0736] 1. Power sensor selection

[0737] Using ABB PGTR23.662A0052 power transmitter

[0738] Measuring range: 0-120% rated power

[0739] Accuracy: ±0.2% full scale

[0740] Response time: <50ms

[0741] 2. Signal acquisition and processing

[0742] Output signal: 4-20mA current signal

[0743] Using 16-bit ADC, the resolution reaches 0.01% full scale

[0744] Sampling frequency: 100Hz

[0745] To achieve anti-aliasing filtering: use a 5th-order Butterworth low-pass filter with a cut-off frequency of 40Hz

[0746] 3. Power signal calibration

[0747] Realize automatic zero point and full scale calibration function

[0748] Linear compensation: Use piecewise linear interpolation to compensate for the nonlinear error of the sensor

[0749] Temperature compensation: Based on the built-in temperature sensor, temperature drift compensation is realized

[0750] 4. Power signal monitoring

[0751] Implementing the power signal validity check algorithm

[0752] Set upper and lower alarm thresholds to detect abnormal power fluctuations

[0753] Power signal mutation detection: based on rate limiting algorithm

[0754] Step 2: PID Controller Design

[0755] 1. PID algorithm implementation

[0756] Using incremental PID algorithm:

[0757] Δu(k)=Kp[e(k)e(k-1)]+Ki*e(k)+Kd[e(k)*2e(k-1)+e(k-2)], where u is the control output, e is the deviation, Kp, Ki, Kd are PID parameters, control period: 10ms

[0758] Use 32-bit floating point operations to ensure calculation accuracy

[0759] 2.PID parameter adaptive adjustment

[0760] Online Adjustment of PID Parameters Based on Fuzzy Logic

[0761] Establishing fuzzy rule base for power deviation and deviation change rate

[0762] Defuzzification using the centroid method

[0763] Parameter adjustment range: Kp (0.5-5.0), Ki (0.1-1.0), Kd (0-0.5)

[0764] 3. Feedforward compensation

[0765] Realize feedforward compensation based on water head changes

[0766] Establish the head-power characteristic curve and use the table lookup interpolation method

[0767] Weighted fusion of feedforward signal and PID output signal

[0768] 4. Anti-integral saturation design

[0769] Adopt integral separation algorithm: when the deviation exceeds the set threshold, the integral action is suspended to implement the integral limiting strategy to prevent system oscillation caused by integral saturation 5. Non-disturbance switching mechanism

[0770] When switching between manual and automatic modes, disturbance-free switching is achieved

[0771] Use tracking-hysteresis algorithm to ensure smooth switching process

[0772] Step 3: Electro-hydraulic conversion

[0773] 1. Proportional valve control

[0774] Moog D633 series proportional valve

[0775] Input signal: ±10V voltage signal

[0776] Realize valve characteristic curve compensation and linearize valve response

[0777] Use PWM technology to drive the proportional valve, frequency 20kHz

[0778] 2. Digital valve control

[0779] High-speed switching valve, response time <3ms

[0780] Adopt pulse width modulation (PWM) control strategy, frequency 200Hz

[0781] Implement dead zone compensation algorithm to improve small signal response performance

[0782] 3. Valve control signal generation

[0783] Generate valve control signal through mapping function based on PID output

[0784] Realize the coordinated control algorithm of proportional valve and digital valve

[0785] Proportional valves are mainly used for small deviations, and digital valves are used for fast response for large deviations.

[0786] 4. Feedback compensation

[0787] Use valve core position feedback signal to build local closed-loop control

[0788] Implement feedforward-feedback composite control strategy to improve dynamic response performance

[0789] Step 4: Actuator Action

[0790] 1. Relay control

[0791] Double-acting hydraulic cylinder, maximum working pressure 16MPa

[0792] Adopt LVDT displacement sensor, resolution 0.01mm

[0793] Realize position closed-loop control, adopt P-PI cascade control structure

[0794] 2. Guide vane opening control

[0795] Establish the corresponding relationship model between the guide vane opening and the servomotor position

[0796] Implement the opening dead zone compensation algorithm to improve the small opening adjustment accuracy

[0797] Adopt adaptive friction compensation strategy to overcome the influence of static friction

[0798] 3. Opening limit

[0799] Realize software and hardware dual opening limit

[0800] Software limitation: Dynamic calculation of maximum allowable opening based on current water head

[0801] Hardware Limits: Setting Mechanical Limits and Hydraulic Locking Devices

[0802] Step 5: Speed ​​Deviation Calculation

[0803] 1. Speed ​​signal processing

[0804] The dual speed measurement scheme described in Section 2.2 is used to obtain the speed signal.

[0805] Implement signal fusion algorithm: Based on the weighted average method, the PT residual pressure frequency measurement and SSG gear frequency measurement signals are fused, and the Kalman filter is used to smooth the signal and filter out high-frequency noise.

[0806] 2. Setting the given speed

[0807] During normal operation, the given speed is the rated speed (such as 300rpm)

[0808] When connected to the grid, the given speed is dynamically adjusted according to the grid frequency.

[0809] Implementing a ramp function generator to smooth changes in the speed setpoint

[0810] 3. Deviation calculation

[0811] Use 32-bit floating point numbers to calculate the speed deviation: e(t) = n_ref(t)n_actual(t)

[0812] Among them, e(t) is the deviation, n_ref(t) is the given speed, and n_actual(t) is the actual speed

[0813] Deviation calculation frequency: 100Hz

[0814] Dead zone processing is achieved: when the deviation is less than ±0.02%, it is considered as zero deviation

[0815] 4. Biased Prediction

[0816] Based on historical deviation data, linear extrapolation is used to predict the deviation prediction results of the next 100ms for feedforward control to improve the system response speed

[0817] Step 6: Speed ​​Controller

[0818] 1. Controller structure

[0819] Adopt cascade PID control structure: outer ring speed PID + inner ring guide vane position PID to realize adaptive PID algorithm, automatically adjust PID parameters according to operating conditions 2. Speed ​​PID controller

[0820] Using position PID algorithm:

[0821] u(t)=Kp*e(t)+Ki∫e(t)dt+Kd*de(t) / dt

[0822] Initial PID parameters: Kp = 2.0, Ki = 0.5, Kd = 0.1

[0823] Realize parameter self-tuning function: online parameter adjustment based on fuzzy rules 3. Guide vane position PID controller

[0824] Adopt incremental PID algorithm to avoid integral saturation problem and implement anti-integral saturation strategy: integral separation + variable integral algorithm

[0825] 4. Operation mode switching

[0826] Island mode: using PIDF (proportional-integral-derivative-feedforward) controller

[0827] * The feedforward term is based on the load change rate to improve the large disturbance response performance

[0828] Grid-connected mode: Slip control

[0829] *Permanent slip rate adjustable range: 2%-8%

[0830] * Realize soft slip characteristics and improve system stability

[0831] 5. Stability optimization

[0832] Implement variable structure control strategy: Use sliding mode control when there is a large deviation, add disturbance observer to improve robustness, estimate and compensate for external disturbances and model uncertainty 6. Oscillation suppression

[0833] Implement adaptive notch filter to suppress the natural oscillation frequency of the electromechanical axis system. Use acceleration feedback control to enhance system damping.

[0834] Step 7: Calculation of guide vane opening

[0835] 1. Characteristic curve modeling

[0836] Establish the turbine characteristic curve model: Q11-n11 characteristic and η-n11 characteristic adopt cubic spline interpolation method to achieve accurate expression of characteristic curve

[0837] 2. Calculation of the optimal operating point

[0838] Based on the current head and load requirements, the optimal guide vane opening and runner blade angle are calculated. The nonlinear equations are solved using the Newton iteration method to determine the optimal operating point.

[0839] 3. Guide vane opening compensation

[0840] Realize water head compensation: dynamically adjust the guide vane opening according to the real-time water head measurement value

[0841] Consider the impact of tailwater level changes and perform secondary compensation

[0842] 4. Opening limit processing

[0843] Realize soft opening limit: dynamically calculate the maximum allowable opening based on the current water head and unit status

[0844] Hard opening limit: set the absolute maximum opening to prevent overload operation

[0845] 5. Opening rate control

[0846] Implement a variable slope limiter to control the rate of change of the guide vane opening

[0847] In the case of large disturbances, appropriately relax the change rate limit to improve the response speed

[0848] Step 8: Perform adjustments

[0849] 1. Electro-hydraulic conversion

[0850] Adopt high performance servo valve, rated flow rate 100L / min, response frequency>50Hz

[0851] Realize valve characteristic curve compensation and linearize valve response

[0852] Adopt PWM drive technology, frequency 20kHz, improve control accuracy

[0853] 2. Guide vane actuator control

[0854] Adopt double closed-loop control: position outer loop + speed inner loop

[0855] Realize adaptive friction compensation to overcome the influence of static friction

[0856] Add acceleration feedforward control to improve dynamic response performance

[0857] 3. Feedback control optimization

[0858] Adopt high-precision LVDT displacement sensor with a resolution of 0.01mm

[0859] Enables dynamic calibration and temperature compensation of sensor signals

[0860] Use Kalman filter to fuse position and velocity information to improve feedback signal quality

[0861] 4.Guide vane synchronization control (applicable to multi-guide vane units)

[0862] Implement guide vane synchronization control algorithm to ensure consistent opening of each guide vane

[0863] The master-slave control structure is adopted to ensure the coordination of the guide vane movement.

[0864] S7: Control actions are performed by coordinated proportional valves and digital valves

[0865] The coordinated control of proportional valves and digital valves can combine the advantages of both valves to achieve both precise and fast control response.

[0866] S7.1: Achieve precise and continuous control of proportional valves

[0867] 1. Use PWM technology to drive the proportional valve, and the frequency is set to 20kHz:

[0868] Using high-frequency PWM signals can reduce the vibration of the valve core and improve control accuracy.

[0869] The PWM duty cycle is linearly related to the valve opening: Opening = k Duty_Cycle + b

[0870] 2. Realize valve characteristic curve compensation:

[0871] Measure and fit the actual valve characteristic curve: Flow = f(Opening)

[0872] Design the inverse compensation function: Opening = f -1 (Desired_Flow)

[0873] 3. Using valve core position feedback signal:

[0874] Using LVDT (Linear Variable Differential Transformer) to measure valve spool position

[0875] Build a local PID control loop to accurately control the valve core position

[0876] S7.2: Realize fast response control of digital valves

[0877] 1. Use pulse width modulation control strategy, and set the frequency to 200Hz:

[0878] Use on-off mode to quickly switch the open and closed state of the digital valve

[0879] Control the average flow rate by adjusting the opening time ratio

[0880] 2. Design a multi-level switching strategy:

[0881] Decompose a large flow digital valve into multiple small valves with binary weights

[0882] For example: valve combinations of 1 / 2, 1 / 4, 1 / 8, 1 / 16 flow rates

[0883] 3. Realize dead zone compensation:

[0884] Identify the opening and closing deadband of digital valves

[0885] Add advance to the control signal to compensate for the dead zone effect

[0886] S7.3: Realize valve characteristic curve compensation

[0887] 1. Establish a mathematical model of valve characteristics:

[0888] Q=Cv f(x)(ΔP / ρ)

[0889] Where Q is the flow rate, Cv is the flow coefficient, f(x) is the opening characteristic function, ΔP is the pressure drop, and ρ is the fluid density

[0890] 2. Design equal percentage characteristic compensation:

[0891] x=k ln(Q / Qmax)+b

[0892] Where x is the valve opening signal after compensation

[0893] 3. Realize adaptive characteristic compensation:

[0894] Online identification of valve characteristic parameters

[0895] Dynamically update compensation function parameters

[0896] S7.4: Building a local closed-loop control

[0897] 1. Use valve core position sensor and flow sensor to build double closed loop control:

[0898] Inner ring: fast response control of valve core position

[0899] Outer ring: precise flow control

[0900] 2. Design state observer:

[0901] Estimation of unmeasured state variables based on Kalman filter

[0902] Improve the robustness and anti-interference ability of control systems

[0903] 3. Realize feedforward-feedback composite control:

[0904] Calculate feedforward control quantities using system models

[0905] Combined with feedback control to eliminate model errors and external disturbances

[0906] Through the coordinated control of proportional valves and digital valves, the system can achieve rapid dynamic response while ensuring control accuracy, effectively improving the speed regulation performance of the hydro-generator set.

[0907] S8: Realize multi-agent collaborative control mechanism

[0908] The multi-agent collaborative control mechanism aims to optimize the operating efficiency of the entire hydropower station and achieve global optimal control by coordinating the operation of multiple hydro-generator sets.

[0909] S8.1: Establish a multi-layer control architecture

[0910] 1. Strategic layer:

[0911] Formulate overall power generation plan for power plants;

[0912] Consider long-term water forecasts, grid demand and economic benefits;

[0913] Output: day-ahead power generation plan, unit start-up and shutdown strategies;

[0914] 2. Tactical layer:

[0915] Coordinate the output distribution of multiple units;

[0916] Consider unit efficiency curve, head changes and load fluctuations;

[0917] Output: power setting value and operation mode of each unit;

[0918] 3. Execution layer:

[0919] Execute specific speed control instructions;

[0920] Achieve precise control of single unit;

[0921] Output: guide vane opening, speed control signal;

[0922] S8.2: Implementing Distributed Optimization Algorithms

[0923] 1. Decentralized global optimization using alternating direction method of multipliers (ADMM):

[0924] Decompose the global optimization problem into multiple local sub-problems;

[0925] Global convergence is achieved by iteratively solving subproblems and updating Lagrange multipliers;

[0926] 2. ADMM algorithm process:

[0927] Initialization: x0, z0, y0

[0928] Iterative solution:

[0929] xi k+1 =argmin(Li(xi,z k ,y k ))

[0930] z k+1 =argmin(ΣLi(xi k+1 ,z,y k ))

[0931] yi k+1 =yi k +ρ(xi k+1 z k+1 )

[0932] Until convergence or the maximum number of iterations is reached

[0933] 3. Design a consistency algorithm based on graph theory:

[0934] Constructing the communication topology between units

[0935] Implementing the average consistency protocol: xi(k+1)=xi(k)+Σj∈Nia ij (x j (k)x i (k))

[0936] Ensure that multiple subsystems achieve consistent control objectives.

[0937] S8.3: Apply game theory methods

[0938] 1. Implement the Nash equilibrium solver:

[0939] A utility function model between units is established, and the best response iteration method is used to solve the Nash equilibrium point.

[0940] 2. Design incentive mechanism:

[0941] Introduce a virtual currency system to reward units with great contributions, and design a scoring rule based on unit efficiency and response speed.

[0942] 3. Implement cooperative game strategy:

[0943] The Shapley value method is used to allocate collaborative benefits, build an alliance formation game model, and optimize the unit collaboration mode.

[0944] S8.4: Establish dynamic communication topology

[0945] 1. Implement adaptive communication strategies:

[0946] Dynamically adjust communication frequency according to system status;

[0947] Increase the communication frequency at critical moments (such as sudden load changes).

[0948] 2. Design a robust information exchange protocol:

[0949] Adopt self-organizing network based on ZigBee.

[0950] Implement multi-path routing to improve communication reliability.

[0951] Use forward error correction coding to deal with communication errors.

[0952] 3. Implement event-triggered communication mechanism:

[0953] Updates are only sent when the state changes by more than a threshold.

[0954] Reduce communication burden and improve system real-time performance.

[0955] S8.5: Implement collaborative learning mechanism

[0956] 1. Adopting a federated learning framework:

[0957] Each unit trains the neural network model locally.

[0958] The central server aggregates model parameters and updates the global model.

[0959] Protect privacy while enabling knowledge sharing.

[0960] 2. Design knowledge distillation mechanism:

[0961] The teacher network is trained to capture the global optimal policy.

[0962] Use the teacher network to guide the student networks of each unit.

[0963] Realize efficient transfer of global knowledge to local models.

[0964] 3. Implement multi-agent reinforcement learning:

[0965] Adopts Actor-Critic architecture.

[0966] Design Q-functions that take into account the actions of other agents.

[0967] Improving learning stability using experience replay and target networks.

[0968] S8.6: Building a global state estimator

[0969] 1. Using the distributed Kalman filter algorithm:

[0970] Each subsystem maintains a local state estimate;

[0971] Update the global state estimate through information fusion;

[0972] Robust design considering communication delay and packet loss;

[0973] 2. Implement anomaly detection and fault isolation mechanisms:

[0974] Anomaly detection using principal component analysis (PCA);

[0975] Support vector machine (SVM) is used to achieve fault classification;

[0976] Design a fault diagnosis algorithm based on particle filtering;

[0977] 3. Build a distributed state observer:

[0978] Design consistency gain using graph Laplacian matrix;

[0979] Ensure convergence to the true state within a finite time;

[0980] S8.7: Optimizing load balancing and energy efficiency

[0981] 1. Implement dynamic load distribution algorithm:

[0982] Real-time optimization based on unit efficiency curve

[0983] Consider the effect of head changes on efficiency

[0984] Using dynamic programming algorithm to solve the optimal load distribution

[0985] 2. Design a predictive maintenance strategy:

[0986] Use historical data to build equipment health models

[0987] Predict the remaining useful life of key components

[0988] Optimize maintenance schedules to minimize impact on overall performance

[0989] 3. Realize economic load dispatch:

[0990] Considering power generation costs, start-up and shutdown costs, and network losses

[0991] Solving the optimal scheduling solution using mixed integer programming

[0992] Combine short-term load forecasting to achieve rolling optimization

[0993] By implementing these multi-agent collaborative control mechanisms, the turbine generator sets can be optimized at a larger system level, improving the operating efficiency and reliability of the entire hydropower station.

[0994] Here is a practical example:

[0995] Scenario description:

[0996] The power station has three hydroelectric generating units, marked as G1, G2 and G3.

[0997] Total installed capacity is 150MW (50MW per unit)

[0998] Upstream reservoir water levels and downstream tailwater levels may change

[0999] Grid load demand changes over time

[1000] The speed regulation steps include:

[1001] 1. Establish a multi-layer control architecture:

[1002] a) Strategic level (power station level):

[1003] Develop a 24-hour power generation plan based on water situation forecasts and grid demand.

[1004] Output: Total electricity generation target per hour.

[1005] b) Tactical level (crew coordination level):

[1006] Coordinate the output distribution of the three units based on strategic level output.

[1007] Consider each unit's efficiency curve, current status and maintenance schedule.

[1008] Output: Power setting value and operating mode of each unit.

[1009] c) Execution layer (single unit level):

[1010] Perform specific speed control.

[1011] Realize precise power control and speed regulation of each unit.

[1012] 2. Implement distributed optimization algorithm:

[1013] Use ADMM algorithm to optimize load distribution:

[1014] Objective function: Maximize the overall power generation efficiency.

[1015] Constraints: The total power generation meets the demand and each unit operates within the allowable range.

[1016] ADMM Iteration Process:

[1017] a) Each unit solves local sub-problems and optimizes its own output.

[1018] b) The central coordinator collects the solutions of each unit and calculates the global optimal solution.

[1019] c) Update the Lagrange multiplier and feed back the global information to each unit.

[1020] d) Repeat the above steps until convergence.

[1021] 3. Apply game theory methods:

[1022] Design a profit distribution mechanism based on Shapley value:

[1023] Calculate the marginal contribution of each unit to the overall efficiency.

[1024] Virtual credit points are allocated based on contribution.

[1025] Credits can be used to adjust maintenance priorities or resource allocations.

[1026] 4. Establish dynamic communication topology:

[1027] Implement a self-organizing network based on ZigBee: Each unit controller acts as a network node. Dynamically adjust the communication frequency, such as increasing the communication frequency when the load changes rapidly. Implement multi-path routing to improve communication reliability.

[1028] 5. Realize collaborative learning mechanism:

[1029] Adopting a federated learning framework: Each unit trains a neural network model locally to predict the optimal control parameters. The central server aggregates model parameters regularly and updates the global model. The updated global model is distributed to each unit to improve the overall control performance.

[1030] 6. Establish a global state estimator:

[1031] Use the distributed Kalman filter algorithm: Each unit maintains a local state estimate (such as speed, power, head, etc.). Units exchange state information and uncertainty estimates. Fusion of multi-source information to build a global system state estimate.

[1032] 7. Optimize load balancing and energy efficiency:

[1033] Implement dynamic load distribution algorithm: monitor the efficiency curve of each unit in real time. When the water head changes, recalculate the optimal load distribution. Transfer the load to the unit with higher efficiency while ensuring the total output.

[1034] Specific operation example:

[1035] 1. Assume that at a certain moment, the grid demand is 120MW, and under the current water head conditions:

[1036] G1 efficiency best point is 45MW

[1037] G2 efficiency best point is 40MW

[1038] G3 efficiency best point is 35MW

[1039] 2. The ADMM algorithm at the tactical layer performs optimization calculations to obtain the optimal allocation:

[1040] G1:48MW

[1041] G2:42MW

[1042] G3:30MW

[1043] 3. Each unit controller at the execution layer receives the power setting value and starts adjusting:

[1044] The unit adjusts the guide vane opening through adaptive PID control

[1045] Federated learning trained model assists in predicting optimal PID parameters

[1046] 4. Dynamic communication system continuously monitors the status of each unit:

[1047] The G3 was found to have slightly higher vibration levels and may require maintenance

[1048] Increased communication frequency of G3 status information

[1049] 5. Game theory based incentive mechanism calculation:

[1050] G1 contributes the most and gets 3 credits

[1051] G2 has the second highest contribution and gets 2 credits

[1052] G3 contributes the least and gets 1 credit point

[1053] 6. Comprehensive analysis of global state estimator:

[1054] It is predicted that the water head may drop in the next 2 hours

[1055] Recommendations for tactical layer to re-optimize load distribution

[1056] Through this example, the embodiment of the present application can use the multi-agent collaborative control mechanism to coordinate the operation of multiple units at different levels, achieve global optimization, and adapt to complex and changing operating environments. This method not only improves the overall efficiency and flexibility of the system, but also enhances the robustness and adaptability of the system.

[1057] Embodiment 2

[1058] Figure 3 The structure diagram of an intelligent speed regulating device for a hydro-generator set provided in an embodiment of the present disclosure is as follows. The intelligent speed regulating device 300 for a hydro-generator set comprises:

[1059] Speed ​​measurement module 31, adopts dual speed measurement scheme of PT residual pressure frequency measurement and SSG gear disc frequency measurement;

[1060] A signal processing module 32, used for fusing and filtering the speed measurement signal;

[1061] The intelligent control module 33 includes:

[1062] Generalizable policy network generator for behavioral cue diffusion,

[1063] YY+242056P

[1064] Computation tree logic verification unit,

[1065] Performance Evaluation and Strategy Selection Unit;

[1066] Adaptive PID controller 34;

[1067] The electro-hydraulic conversion module 35 includes a proportional valve and a digital valve for coordinated control;

[1068] Actuator 36, used to adjust the guide vane opening;

[1069] The intelligent control module 33 generates, verifies and selects the optimal control strategy according to the system status and control objectives, and realizes efficient speed control of the hydro-generator set through the adaptive PID controller and the electro-hydraulic conversion module.

[1070] The device of the embodiments of the present disclosure can execute the method provided by the embodiments of the present disclosure, and the implementation principles are similar. The actions performed by each module in the device of each embodiment of the present disclosure correspond to the steps in the method of each embodiment of the present disclosure. For the detailed functional description of each module of the device, please refer to the description in the corresponding method shown in the previous text, which will not be repeated here.

[1071] The above is only an optional implementation method for some implementation scenarios of the present disclosure. It should be pointed out that for ordinary technicians in this technical field, without departing from the technical concept of the scheme of the present disclosure, other similar implementation methods based on the technical ideas of the present disclosure are also within the protection scope of the embodiments of the present disclosure.

Claims

1. An intelligent speed regulation method for a hydro-generator set, characterized in that: The following steps are involved: A dual speed measurement scheme is used to obtain the unit speed signal, and the dual speed measurement scheme includes PT residual pressure frequency measurement and SSG gear disc frequency measurement; Performing fusion processing and filtering on the speed signal; Based on the processed speed signal, a generalizable strategy network generator using behavioral cue diffusion is used to generate a candidate control strategy; Use computational tree logic to verify the security and effectiveness of the generated policies; Performing performance evaluation on the strategies that have passed the verification and selecting the optimal strategy; Applying the selected optimal strategy to the adaptive PID controller; The control action is executed through the coordinated control of the proportional valve and the digital valve to adjust the opening of the turbine guide vane and realize the speed control of the turbine generator set.

2. The method according to claim 1, characterized in that In the dual speed measurement scheme: The PT residual voltage frequency measurement uses a voltage transformer to obtain a frequency signal, and the frequency measurement accuracy is 0.001Hz; The SSG gear frequency measurement installs a gear on the shaft of the turbine generator set and uses an electromagnetic proximity switch to generate a frequency signal. The frequency measurement range is 0.2Hz-150Hz.

3. The method according to claim 1, characterized in that The fusion processing and filtering of the speed signal include: The weighted average method is used to fuse the PT residual pressure frequency measurement and SSG gear frequency measurement signals; The Kalman filter is used to smooth the fused signal and filter out high-frequency noise.

4. The method according to claim 1, characterized in that The generalizable policy network generator for behavior cue diffusion includes the following steps: A behavior prompt space including discrete and continuous behavior prompts is defined, wherein the discrete behavior prompts include stable operation, fast response and energy-saving mode, and the continuous behavior prompts include target speed and expected response time; Build a strategic network architecture, including: An encoder network consisting of a multi-layer perceptron and a one-dimensional convolutional neural network is used to map system states and behavior cues into a latent space. A decoder network with a variational autoencoder structure is used to generate control policy parameters from the latent space. Implement multi-head attention mechanisms in encoders and decoders to dynamically focus on relevant state and behavior cues; Design a noise injection scheme to inject Gaussian noise into the latent space and implement adaptive noise scaling; Through the denoising network of the U-Net structure, the gradual denoising process from random strategy to target strategy is realized; The policy network is trained using a multi-objective loss function consisting of reconstruction error, KL divergence, behavioral consistency loss, and adversarial loss. Priority experience replay and importance sampling techniques are used to improve training efficiency.

5. The method according to claim 1, characterized in that Verifying the security and effectiveness of the generated policy using computational tree logic includes the following steps: Define atomic propositions that describe the key states and control actions of the system, including speed state, power state and control actions; Construct a temporal logic formula that represents the system characteristics, including: Safety property formula: AG(speed≤max_speed), where speed is the current speed and max_speed is the maximum allowable speed; Active quality formula: AF[0,T](|speed target_speed|≤ε), where target_speed is the target speed, T is the predetermined time, and ε is the allowable error; Stability property formula: AG(AF[0,T](|speed nominal_speed|≤δ)), where nominal_speed is the rated speed and δ is the steady-state error range; Establish a state transition diagram to represent the speed control system, and use the adjacency matrix to represent the transition relationship between states; Implement the depth-first search algorithm, combine memoized search and bidirectional search strategies, and verify the above temporal logic formula; Design a counterexample generation mechanism to provide a specific state sequence as a counterexample when verification fails; Implement a lightweight online checker to continuously monitor whether key attributes satisfy the above logical formula; Adopting incremental verification algorithm, only verifying the part of strategy changes, improving real-time verification speed; State space representation and verification efficiency are optimized using symbolic model checking techniques including binary decision diagrams (BDDs).

6. The method according to claim 1, characterized in that Methods for evaluating the performance of validated strategies include: Define multi-dimensional evaluation indicators, including stability indicators, dynamic performance indicators, energy efficiency indicators and robustness indicators; Use turbine-generator model to conduct closed-loop simulation to simulate various operating conditions and disturbance scenarios; The weighted sum method is used to synthesize various indicators, and the Pareto optimal selection algorithm is used to balance multiple objectives.

7. The method according to claim 1, characterized in that The parameter adjustment mechanism of the adaptive PID controller includes: Based on fuzzy logic, PID parameters are adjusted online to establish a fuzzy rule base for power deviation and deviation change rate; Use the centroid method to perform defuzzification operations and dynamically adjust the Kp, Ki, and Kd parameters; Implement the integral separation algorithm and integral limiting strategy to prevent system oscillation caused by integral saturation.

8. The method according to claim 1, characterized in that The coordinated control strategy of proportional valves and digital valves includes: Use proportional valve to achieve precise continuous control, driven by PWM technology with a frequency of 20kHz; A digital valve is used to achieve fast response, and a pulse width modulation control strategy is adopted with a frequency of 200Hz; When the deviation is less than the preset threshold, a proportional valve is used, and when the deviation is greater than the preset threshold, a digital valve is used for quick response; Realize valve characteristic curve compensation and linearize valve response; The valve core position feedback signal is used to build a local closed-loop control.

9. The method according to claim 1, characterized in that: The method also includes implementing a multi-agent collaborative control mechanism, and the specific steps are as follows: Establish a multi-layered control architecture, including strategic, tactical and executive layers, including: The strategic level includes developing the overall power generation plan for the power station; The tactical layer includes coordinating the output distribution of multiple units; The execution layer includes executing specific speed control instructions; Implement distributed optimization algorithms, including: Decentralized global optimization using alternating direction multiplier method ADMM; Design a consistency algorithm based on graph theory to ensure the coordinated operation of multiple subsystems; Apply game theory methods, including: Implement a Nash equilibrium solver to find optimal strategies in competitive and cooperative scenarios; Design incentive mechanisms to promote effective collaboration among units; Establish dynamic communication topology, including: Implement adaptive communication strategies to dynamically adjust communication frequency and content based on system status; Design robust information exchange protocols to cope with communication delays and packet loss; Implement collaborative learning mechanisms, including: Adopting a federated learning framework to achieve knowledge sharing while protecting privacy; Design a knowledge distillation mechanism to effectively transfer the global optimal strategy to each subsystem; Build a global state estimator, including: Use the distributed Kalman filter algorithm to fuse multi-source information and construct the global system state; Implement anomaly detection and fault isolation mechanisms to improve the overall robustness of the system; Optimize load balancing and energy efficiency, including: Implement dynamic load distribution algorithm to optimize overall energy efficiency according to the efficiency curve of each unit; Design predictive maintenance strategies and coordinate maintenance plans for multiple units to minimize the impact on overall performance.

10. An intelligent speed regulating device for a hydro-generator set, characterized in that: include: The speed measurement module adopts a dual speed measurement scheme of PT residual pressure frequency measurement and SSG gear disc frequency measurement; A signal processing module, used for fusing and filtering the speed measurement signal; Intelligent control module, including: Generalizable policy network generator for behavioral cue diffusion, Computation tree logic verification unit, Performance Evaluation and Strategy Selection Unit; YY+242056P Adaptive PID controller; Electro-hydraulic conversion module, including proportional valves and digital valves for coordinated control; Actuator, used to adjust the guide vane opening; The intelligent control module generates, verifies and selects the optimal control strategy according to the system status and control objectives, and realizes efficient speed control of the hydro-generator set through the adaptive PID controller and the electro-hydraulic conversion module.

Citation Information

Cited By

  • Generator speed measurement braking system

    CN119226724A

  • Generator speed measuring brake system

    CN119226724B

  • Cooperative task-oriented aircraft meta-model autonomous learning control system

    CN119250125A

  • A control system for autonomous learning of aircraft meta-model for collaborative tasks

    CN119250125B

  • Hydraulic servo system for water turbine and control device of hydraulic servo system

    CN119467449A