An intelligent start-stop system and method for hydroelectric generating units based on multi-dimensional dynamic optimization and high-precision cooperative control

Abstract

The application discloses a kind of based on multi-dimension dynamic optimization and high-precision collaborative control's hydroelectric generating set intelligent start-stop system and method, it is related to the technical field of intelligent control of hydropower plant, the system includes: data acquisition module, for real-time monitoring hydroelectric generating set state and acquisition operating data;Dynamic priority management module, for based on water head, vibration area, fault signal and historical operation data, dynamically generate priority weight, realize hydroelectric generating set start sequence global optimization;Collaborative interpolation control module, for converting dispatching plan curve into control instruction, and introduce dynamic error threshold check mechanism, trigger multi-stage alarm and optimize response speed;V-shaped curve identification module;Intelligent diagnosis module;Risk scoring module.The application realizes the comprehensive promotion of unit start-stop success rate, regulation response speed and fault positioning accuracy under complex working conditions through multi-module collaborative closed loop, provides safe and efficient flexible regulation support for high proportion new energy power grid.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology for hydropower plants, and in particular to an intelligent start-up and shutdown system and method for hydropower units based on multi-dimensional dynamic optimization and high-precision collaborative control. Background Technology

[0002] Against the backdrop of accelerated construction of new power systems, hydropower units, due to their rapid response and flexible adjustment characteristics, have become a core support for power grid peak shaving and frequency regulation. Especially with the continuous increase in the proportion of new energy sources such as wind and solar power, large hydropower units need to be frequently started and stopped to balance grid fluctuations, and the success rate of start-up and shutdown directly affects the safety and economy of the power station. Traditional hydropower unit start-up and shutdown systems generally adopt a "manually preset priority + fixed threshold control" mode: operators set the start-up and shutdown sequence based on static parameters (such as unit rated capacity and historical failure rate), the dispatching system issues 15-minute or 5-minute load curves, and then on-site personnel operate step-by-step through the monitoring system. This mode was applicable in the early stages of small-scale grid connection, but facing the complex operating conditions brought about by a high proportion of new energy access, it has gradually exposed multiple technical bottlenecks, manifested in the following aspects:

[0003] 1. Static Priority: Relying on manually preset priorities, it cannot dynamically adapt to changes in head, vibration zone shifts, and fluctuations in unit health status. The static priority mechanism is difficult to adapt to dynamic operating conditions; when head fluctuates, vibration zone shifts, or unit health status changes, the preset priorities cannot be adjusted in real time. For example, high-head impulse turbines are prone to entering the vibration zone in low-load ranges. Traditional systems still start and stop in a fixed sequence, which aggravates the mechanical stress of the equipment and may even induce overspeed protection malfunctions.

[0004] 2. Coarse-grained control commands: The 15-minute or 5-minute planning curves issued by the scheduling system have limited interpolation accuracy and are difficult to be directly applied to real-time control at the second level, thus causing output deviation.

[0005] 3. Poor adaptability to special scenarios: The V-curve period is prone to accidental shutdown, and there is a lack of rapid response mechanism when the load changes suddenly; for example, when the technical water supply pressure is abnormal or the spindle seal fails, the system only triggers a general alarm and does not adjust the start-stop sequence in conjunction with it; emergency shutdown relies on a single path of mechanical overspeed protection. If the speed measuring device is aging (such as the gear plate sensor fails), it may delay the best time to block the jet and exacerbate the risk of water hammer.

[0006] 4. Vague fault diagnosis: The alarm information is too general and fails to provide precise fault location and corresponding repair suggestions at the equipment level; in addition, protections such as oil flow interruption and bearing temperature use single-point sensors, which have a high false trigger rate and cannot distinguish between mechanical wear and instantaneous interference.

[0007] These bottlenecks limit the regulatory value of hydropower in new power systems, and there is an urgent need to achieve a technological leap through multi-dimensional dynamic optimization and high-precision collaborative control. Summary of the Invention

[0008] Therefore, it is necessary to provide an intelligent start-up and shutdown system and method for hydropower units based on multi-dimensional dynamic optimization and high-precision collaborative control to address the above-mentioned technical problems.

[0009] In a first aspect, the present invention provides an intelligent start-up and shutdown system for hydropower units based on multi-dimensional dynamic optimization and high-precision collaborative control, the system comprising:

[0010] The data acquisition module is used to monitor the status of the hydropower unit and collect operating data in real time;

[0011] The dynamic priority management module is used to dynamically generate priority weights based on head, vibration zone, fault signals and historical operating data, so as to achieve global optimization of the hydropower unit startup sequence.

[0012] The collaborative interpolation control module is used to convert scheduling plan curves into control commands and introduces a dynamic error threshold verification mechanism to trigger multi-level alarms and optimize response speed.

[0013] The V-curve recognition module is used to construct spatiotemporal composite V-curve recognition rules and load change inertial transition strategies to avoid equipment losses and oscillation risks caused by the start-up and shutdown of hydropower units.

[0014] The intelligent diagnostic module is used to build a multi-level fault diagnosis framework to achieve fault location and self-healing.

[0015] The risk scoring module is used to calculate risk priority numbers based on failure mode analysis of key components of hydropower units and to mark them as fault risk scores integrated into the dynamic priority weight calculation model.

[0016] Furthermore, the dynamic priority management module includes:

[0017] The priority weight calculation module is used to integrate water head, vibration zone, fault signals and historical operating data to build a weight calculation model and calculate the priority weight at the current moment.

[0018] The vibration zone self-learning module is used to adaptively update the vibration zone boundary based on historical operating data, and feed the updated vibration zone range back to the start-up and shutdown control logic to dynamically optimize the crossing strategy.

[0019] The start-stop sequence optimization module is used to optimize the start-stop sequence of hydropower units using the Hungarian algorithm to maintain the overall operating efficiency of the hydropower units at the target level.

[0020] Furthermore, the vibration zone self-learning module includes:

[0021] The vibration acquisition module is used to collect operating data of the hydropower unit, including speed, power, vibration value, speed changes during start-up and shutdown, and abnormal vibration characteristics during start-up and shutdown.

[0022] The initial determination module is used to set the initial vibration zone range, which serves as the initial input for the vibration zone model.

[0023] The annotation and training module is used to perform offline analysis of historical operating data, extract the rotational speed range of abnormal vibration points, statistically analyze the high vibration range under similar working conditions, and use machine learning to build a model to generate a vibration zone model. It also extracts and outputs the target vibration frequency band through cluster analysis.

[0024] The online self-learning module is used to introduce a sliding window algorithm to update the vibration zone range in real time, feed the updated vibration zone range back to the start-up and shutdown control logic, and perform safety fault tolerance verification.

[0025] Furthermore, the introduction of a sliding window algorithm to update the vibration zone range in real time, feeding the updated vibration zone range back to the start-up and shutdown control logic, and performing safety fault tolerance verification includes:

[0026] Set the parameters and data fields for the sliding window;

[0027] Acquire real-time hydropower unit operation data, extract vibration data, load and speed, add new data points to a sliding window, and remove old data that exceeds the constant window length;

[0028] Calculate the mean and variance of the vibration value corresponding to each load segment within the sliding window, detect whether there is an abnormal vibration interval exceeding the preset threshold and preset duration, if so, mark it as an abnormal vibration segment, and dynamically adjust it using the exponential weighted moving average method;

[0029] The number of abnormal markings for load segments is counted. If the number of consecutive markings of a load segment as an abnormal vibration segment exceeds a preset threshold, the abnormal vibration segment is written into the vibration zone model cache.

[0030] The new vibration zone model is compared with the original vibration zone model. If the model deviation exceeds the set threshold, the replacement is temporarily suspended. If the model deviation is within the preset range, passes multiple consecutive verifications, and passes the working condition consistency verification, the new vibration zone model replaces the original vibration zone model.

[0031] The updated vibration zone model's vibration zone range is fed back into the start-up and shutdown control logic to dynamically optimize the crossing strategy and modify the vibration risk item in the start-up and shutdown priority scoring.

[0032] Furthermore, the dynamic adjustment using the exponentially weighted moving average method includes:

[0033] The vibration value of the vibration zone model at the current moment is collected, combined with the smoothed output value of the previous moment, and the smoothed output value at the current moment is calculated using the exponentially weighted average working hours.

[0034] The smoothed output value is used as the basis for determining the new vibration zone, and the model boundary of the vibration zone model is updated. If the real-time vibration value is higher than the updated model boundary for more than a preset time, the control strategy is adjusted.

[0035] Furthermore, the start / stop sequence optimization module includes:

[0036] The initialization module is used to load the basic parameters of the hydropower units and the time window and time slot division of the current scheduling plan. At the beginning of each control cycle, it obtains the real-time priority weight of each hydropower unit.

[0037] The matrix construction module is used to set the set of hydropower units and the set of sequence positions to be assigned, forming sequence pairs of units and sequence positions. For each sequence pair, an allocation cost is set to form a cost matrix.

[0038] The optimal matching module is used to take the cost matrix as input, call the Hungarian algorithm, and obtain a set of one-to-one matching that minimizes the total cost.

[0039] The generation and distribution module is used to arrange the units corresponding to each sequence position in sequence according to the matching result of minimizing the total cost, obtain the optimal start-stop sequence of the current control cycle, and convert the current sequence into start-stop commands, which are then distributed to the local control units of each hydropower unit for execution.

[0040] The iterative feedback module is used to repeatedly perform sequence optimization operations within a preset iteration cycle, dynamically respond to changes in head, vibration zone shifts and fault conditions, and record the running effect and deviation of each match. The adaptive adjustment coefficient in the weight calculation model is optimized synchronously through logs and reinforcement learning.

[0041] Furthermore, the collaborative interpolation control module includes:

[0042] The power output acquisition module is used to set interval time points and periodically acquire the scheduling plan curve according to the interval time points, which represents the planned power output of a hydropower unit;

[0043] The regression analysis module estimates the derivative based on the historical operating data of the hydropower unit and substitutes it into the Hermite interpolation to generate control points every minute. If the difference between the interpolation point and the actual operating data is greater than the preset threshold, an alarm is triggered and the system switches to the backup interpolation method.

[0044] The control execution module is used to transmit control points generated by minute interpolation to the local control unit between two adjacent original time intervals.

[0045] Furthermore, the risk scoring module includes:

[0046] Failure Mode Identification (FMO) module is used to organize all failure modes of hydropower unit components;

[0047] The impact and consequences assessment module is used to assess the impact of each failure mode on the operation of the hydropower unit;

[0048] The probability assessment module is used to score the frequency of occurrence of each failure mode based on historical operating data and hydropower unit failure statistics.

[0049] The detection difficulty assessment module is used to score the difficulty of detecting each failure mode based on existing diagnostic methods;

[0050] The risk priority number calculation module is used to calculate the risk priority number for each failure mode.

[0051] The normalization aggregation module is used to set the maximum value of the risk priority number, normalize the risk priority number of each failure mode with the maximum value of the risk priority number, and perform a weighted average of the normalized risk priority numbers of all failure modes of the same hydropower unit to obtain a comprehensive failure risk score.

[0052] The priority weighting module is used to replace the corresponding item in the weighting calculation model with the comprehensive fault risk score, and to dynamically optimize the output results of the priority weighting of hydropower units.

[0053] The feedback closed-loop optimization module is used to update the maximum value of the risk priority number based on the trend of risk priority number changes, and optimize the evaluation criteria and weight parameters through a self-learning mechanism.

[0054] Furthermore, the process of updating the maximum value of the risk priority number based on its changing trend, and optimizing the evaluation criteria and weight parameters through a self-learning mechanism, includes:

[0055] According to the control cycle, the risk priority number of each failure mode of each hydropower unit is recorded periodically to construct a time series of risk priority numbers;

[0056] Calculate the rate of change of risk priority number for any failure mode over two consecutive control cycles, compare it with a preset rate of change threshold, and determine whether it is increasing, stable, or decreasing.

[0057] If the rate of change of risk priority number is increasing, it is determined that the diagnostic function has not controlled the risk in a timely manner, and the severity or frequency score of the current failure mode is increased; if the rate of change of risk priority number is decreasing, the severity or frequency score of the current failure mode is decreased.

[0058] Assess the impact of current failure mode parameters on the overall risk priority number, set minimizing the error as the objective function, and optimize the parameter combination to maintain the matching effect between the overall priority model score and the actual operating effect;

[0059] After the fault risk score is completed for each control cycle, the optimization results are written into the database.

[0060] Secondly, the present invention also provides an intelligent start-up and shutdown method for hydropower units based on multi-dimensional dynamic optimization and high-precision collaborative control, the method comprising:

[0061] S1. Real-time monitoring of the status of hydropower units and collection of operating data;

[0062] S2. Based on head, vibration zone, fault signals and historical operating data, priority weights are dynamically generated to achieve global optimization of the hydropower unit startup sequence;

[0063] S3. Convert the scheduling plan curve into control commands and introduce a dynamic error threshold verification mechanism to trigger multi-level alarms and optimize response speed;

[0064] S4. Construct spatiotemporal composite V-shaped curve identification rules and load change inertial transition strategies to avoid equipment losses and oscillation risks caused by the start-up and shutdown of hydropower units;

[0065] S5. Construct a multi-level fault diagnosis framework to achieve fault location and self-healing;

[0066] S6. Based on the failure mode analysis of key components of hydropower units, calculate the risk priority number and mark it as a fault risk score to be integrated into the dynamic priority weight calculation model.

[0067] The beneficial effects of this invention are as follows: By deeply integrating dynamic priority optimization, high-precision command coordination, and intelligent diagnostic mechanisms, the regulatory value of hydropower in the new power system is significantly enhanced. It breaks through the limitations of traditional static priority, dynamically adjusting the start-up and shutdown sequence based on real-time head changes, vibration zone offset, and unit health status, thus avoiding equipment damage caused by entering the vibration zone under high load conditions. For coarse-grained scheduling curves, a piecewise cubic Hermite interpolation algorithm is used to generate second-level control commands, coupled with the coordinated timing of the governor, excitation system, and synchronizing device. During sudden load changes, dynamic error verification quickly corrects output deviations, achieving… Precise matching with grid frequency fluctuations; the introduction of a spatiotemporal composite V-shaped curve recognition mechanism can predict load troughs caused by sharp drops in renewable energy output, maintaining the minimum technical output of the unit rather than frequent start-ups and shutdowns, effectively reducing mechanical stress on equipment; a multi-level fault diagnosis framework locates general alarms to specific equipment components, and combined with a self-healing recommendation mechanism, quickly switches to backup equipment or generates maintenance work orders, significantly reducing the risk of unplanned downtime; finally, through multi-module collaborative closed-loop, it achieves a comprehensive improvement in unit start-up and shutdown success rate, regulation response speed, and fault location accuracy under complex operating conditions, providing safe, efficient, and flexible regulation support for high-proportion renewable energy grids. Attached Figure Description

[0068] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings:

[0069] Figure 1 This is a schematic diagram of a smart start-stop system for hydropower units based on multi-dimensional dynamic optimization and high-precision collaborative control according to an embodiment of the present invention.

[0070] Figure 2 This is a flowchart of an intelligent start-up and shutdown method for hydropower units based on multi-dimensional dynamic optimization and high-precision collaborative control according to an embodiment of the present invention;

[0071] Figure 3 This is an architecture diagram illustrating the hardware layer, software modules, and data flow according to an embodiment of the present invention.

[0072] The reference numerals are as follows: 1. Data acquisition module; 2. Dynamic priority management module; 3. Collaborative interpolation control module; 4. V-curve recognition module; 5. Intelligent diagnosis module; 6. Risk scoring module. Detailed Implementation

[0073] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0074] Please see Figure 1 This paper provides an intelligent start-up and shutdown system for hydropower units based on multi-dimensional dynamic optimization and high-precision collaborative control, including: a data acquisition module 1, a dynamic priority management module 2, a collaborative interpolation control module 3, a V-curve recognition module 4, an intelligent diagnosis module 5, and a risk scoring module 6.

[0075] Data acquisition module 1 is used to monitor the status of hydropower units in real time and collect operating data.

[0076] Specifically, data acquisition module 1, as the hardware layer of the system architecture, is deployed on the high-availability server cluster of the central control center, supporting two-way real-time communication with the SCADA system (data acquisition and monitoring control system) and the dispatch master station.

[0077] Data acquisition module 1 achieves second-level monitoring, covering the output P of the hydropower unit. i (t), frequency f(t), vibration zone state S v (t), head H(t), fault signal F i (t) and other parameters.

[0078] The dynamic priority management module 2 is used to dynamically generate priority weights based on water head, vibration zone, fault signals and historical operating data, so as to realize global optimization of the hydropower unit start-up sequence.

[0079] In the description of this invention, the dynamic priority management module 2 includes: a priority weight calculation module, a vibration zone self-learning module, and a start-stop sequence optimization module.

[0080] The priority weight calculation module is used to integrate water head, vibration zone, fault signals and historical operating data to build a weight calculation model and calculate the priority weight at the current moment.

[0081] Specifically, priority weight W i The formula for calculating (t) is:

[0082] ;

[0083] In the formula, α, β, γ, and δ are adaptive adjustment coefficients, which are dynamically optimized using the gradient descent method;

[0084] H max The current upper limit of water head, and T is the historical data time window (default 30 minutes);

[0085] S v (t)∈[0,1] represents the deviation of the vibration zone (0 is safe, 1 is critical state);

[0086] P i(τ) represents the actual output value of the unit over time (unit: MW);

[0087] dτ represents a small element with respect to the time variable τ.

[0088] The vibration zone self-learning module is used to adaptively update the vibration zone boundary based on historical operating data, and feed the updated vibration zone range back to the start-up and shutdown control logic to dynamically optimize the crossing strategy; thus, it can adaptively update the vibration zone boundary based on long-term data and improve the model accuracy.

[0089] In the description of this invention, the vibration zone self-learning module includes: a vibration acquisition module, an initial judgment module, a labeling training module, and an online self-learning module.

[0090] The vibration acquisition module is used to collect operating data of the hydropower unit, including speed, power, vibration value, speed changes during start-up and shutdown, and abnormal vibration characteristics during unit start-up and shutdown.

[0091] Specifically, the vibration acquisition module collects real-time operating data of the unit, especially: speed, power, vibration value (three-dimensional acceleration or velocity sensor), speed changes during start-up and shutdown, and abnormal vibration characteristics during unit start-up and shutdown.

[0092] The initial determination module is used to set the initial vibration zone range, which serves as the initial input for the vibration zone model.

[0093] Specifically, based on historical experience or parameters provided by the manufacturer, an initial vibration zone range is given; a "suspicious vibration zone" is set as the initial input for the model.

[0094] The annotation and training module is used to perform offline analysis of historical operating data, extract the rotational speed range of abnormal vibration points, statistically analyze the high vibration range under similar working conditions, and use machine learning to build a vibration zone model. It also extracts and outputs the target vibration frequency band through cluster analysis.

[0095] Specifically, offline analysis is performed on historical operating data, including extracting the rotational speed ranges of abnormal vibration points, statistically analyzing the high vibration intervals that frequently occur under similar operating conditions, and using machine learning (such as clustering, decision trees, etc.) to perform preliminary modeling, generate vibration zone models, and output the most likely vibration frequency bands.

[0096] For the modeling method of vibration zone model, cluster analysis can be used to extract similar high vibration running sections, and decision trees can be used to construct vibration zone judgment rules. Specific application methods include: analyzing historical vibration data during offline training; combining real-time data during online operation, and dynamically adjusting the vibration zone boundary with sliding window and exponential weighted average (EWMA); the model will continuously learn and adaptively optimize the crossing strategy, such as early acceleration, delayed start and stop, etc.

[0097] This allows the system to more accurately identify and avoid "dangerous sections," achieving safer and more efficient start-stop control.

[0098] The online self-learning module is used to introduce a sliding window algorithm to update the vibration zone range in real time, feed the updated vibration zone range back to the start-up and shutdown control logic, and perform safety fault tolerance verification.

[0099] In the description of this invention, a sliding window algorithm is introduced to update the vibration zone range in real time, and the updated vibration zone range is fed back to the start-up and shutdown control logic, and safety fault tolerance verification is performed, including:

[0100] Step S21: Set the sliding window parameters and data fields.

[0101] Specifically, the sliding window size can be set: such as 30 minutes or 1800 sampling points (calculated at a sampling frequency of 1 second);

[0102] Set data fields: vibration value (3 axes), rotational speed, output, head, etc.

[0103] Step S22: Obtain real-time collected hydropower unit operation data, extract vibration data, load and speed, add new data points to the sliding window, and remove old data that exceeds the constant window length.

[0104] Specifically, the vibration data, load, and speed of the unit are collected in real time from SCADA; new data points are added to the sliding window, and the oldest data is dequeued to keep the window length constant.

[0105] Step S23: Calculate the mean and variance of the vibration value corresponding to each load segment within the sliding window, detect whether there is an abnormal vibration interval exceeding the preset threshold and preset duration, if so, mark it as an abnormal vibration segment, and dynamically adjust it using the exponential weighted moving average method.

[0106] Specifically, it is necessary to calculate the mean and variance of the vibration values ​​corresponding to each load segment within the window; check whether there are abnormal vibration segments that continuously exceed the threshold; and mark these segments as "abnormal vibration segments" or "suspicious vibration segments".

[0107] In the description of this invention, dynamic adjustment using the exponentially weighted moving average method includes:

[0108] Step S231: Collect the vibration value of the vibration zone model at the current moment, combine it with the smoothed output value of the previous moment, and calculate the smoothed output value at the current moment using the exponentially weighted average working hours.

[0109] Specifically, the formula for calculating the smoothed output value at the current moment is:

[0110] ;

[0111] In the formula, α is the smoothing coefficient, and V prev The previous smoothed value; V new The current smoothed output is used to replace the original, highly volatile values; V t This represents the vibration value at the current moment.

[0112] Step S232: Use the smoothed output value as the basis for determining the new vibration zone, update the model boundary of the vibration zone model, and if the real-time vibration value is higher than the updated model boundary for more than a preset time, adjust the control strategy, such as delaying startup or slowing down acceleration.

[0113] Step S24: Count the number of abnormal markings of the load segment. If the number of consecutive markings of a load segment as an abnormal vibration segment is greater than the preset threshold, then write the abnormal vibration segment into the vibration zone model cache.

[0114] For example, if a load segment is marked as a high vibration segment N times consecutively, it is written into the vibration zone model cache.

[0115] Step S25: Compare the new vibration zone model with the original vibration zone model. If the model deviation exceeds the set threshold, the replacement is temporarily suspended. If the model deviation is within the preset range, passes multiple consecutive verifications, and passes the working condition consistency verification, the new vibration zone model replaces the original vibration zone model.

[0116] Specifically, the differences between the new vibration zone model and the original vibration zone model are compared. If the differences exceed a set threshold, the replacement is temporarily suspended.

[0117] The new vibration zone model must meet the following requirements: pass multiple consecutive verifications; consistent operating conditions, such as stable head and output; and historical model deviations within the allowable range.

[0118] Step S26: Feed back the vibration zone range of the updated vibration zone model to the start-stop control logic, dynamically optimize the crossing strategy, and modify the vibration risk item in the start-stop priority score.

[0119] Specifically, the updated vibration zone range will be fed back into the start-stop control logic to automatically avoid the resonance frequency band; adjust the acceleration and excitation strategies; and modify the "vibration risk item" in the start-stop priority score.

[0120] The start-stop sequence optimization module is used to optimize the start-stop sequence of hydropower units using the Hungarian algorithm to maintain the overall operating efficiency of the hydropower units at the target level.

[0121] In the description of this invention, the start-stop sequence optimization module includes: an initialization module, a matrix construction module, an optimal matching module, a generation and distribution module, and an iterative feedback module.

[0122] The initialization module is used to load the basic parameters of the hydropower units and the time window and time slot division of the current scheduling plan. At the beginning of each control cycle, it obtains the real-time priority weight of each hydropower unit.

[0123] Specifically, the system loads all basic parameters of the generating units, including rated capacity, vibration zone range, head efficiency curve, and the time window and time slot division of the current scheduling plan. At the beginning of each control cycle, the system obtains the real-time priority weights of each generating unit, calculated based on factors such as comprehensive head, vibration zone deviation, fault signals, and historical output contribution.

[0124] The matrix construction module is used to set the set of hydropower units and the set of sequence positions to be assigned, forming sequence pairs of units and sequence positions. For each sequence pair, an allocation cost is set to form a cost matrix.

[0125] Specifically, let the set of generating units be... The set of sequence positions to be assigned (i.e., the order of the 1st to Nth start / stop operations) is: .

[0126] For each pair Define allocation cost: .

[0127] Among them, W i For unit u i Dynamic priority weights; f slot (j) represents the sequence position s j The "position weight" can be defined, for example, as a function that decreases linearly or exponentially with increasing sequence number, to reflect that "the earlier the start and stop, the better the overall efficiency".

[0128] The optimal matching module takes the cost matrix as input, calls the Hungarian algorithm, and obtains a set of one-to-one matchings that minimize the total cost.

[0129] Specifically, the above Cost matrix As input, the classic Munkres algorithm is invoked to obtain a set of one-to-one matching pairs that minimize the total cost:

[0130] ;

[0131] Algorithm Time Complexity Generally, this is to meet the scheduling requirements of this system at the second or minute level.

[0132] The generation and distribution module is used to arrange the units corresponding to each sequence position in sequence according to the matching result of minimizing the total cost, obtain the optimal start-stop sequence for the current control cycle, and convert the current sequence into start-stop commands, which are then distributed to the local control units of each hydropower unit for execution.

[0133] Specifically, based on the matching results, each sequence position s j Corresponding unit u i Arranged sequentially, the optimal start-stop sequence for this cycle is obtained:

[0134] ;

[0135] To convert this sequence into specific start / stop commands, it is necessary to combine it with minute-level control commands generated by the collaborative interpolation module and send them to the LCU (Local Control Unit) of each unit for execution.

[0136] The iterative feedback module is used to repeatedly perform sequence optimization operations within a preset iteration cycle, dynamically respond to changes in head, vibration zone shifts and fault conditions, and record the running effect and deviation of each match. The adaptive adjustment coefficient in the weight calculation model is optimized synchronously through logs and reinforcement learning.

[0137] Specifically, in each predefined iteration cycle, such as every minute or when there is a sudden change in load, the above modules (initialization module, matrix construction module, optimal matching module, and generation and distribution module) are re-executed to dynamically respond to changes in head, vibration zone shifts, or fault conditions.

[0138] Record the running effect and deviation of each matching, and continuously optimize the adaptive coefficients in the weight calculation through logs and reinforcement learning modules, such as optimizing α, β, γ, and δ using gradient descent.

[0139] The collaborative interpolation control module 3 is used to convert the scheduling plan curve into control commands and introduce a dynamic error threshold verification mechanism to trigger multi-level alarms and optimize response speed.

[0140] Specifically, the original scheduling plan time points are the original daily plan curve issued by the scheduling system, which consists of 96 points per day, one point every 15 minutes. The program calculates this to one point per minute using interpolation; it is represented by the subscript j and denoted as t. j This corresponds to the 15-minute interval specified in the scheduling document, such as: (15-minute interval).

[0141] The control time point generated by interpolation is represented by the subscript k and denoted as t. k This corresponds to a 1-minute interval time point generated by the system, such as: (1 minute interval).

[0142] Adaptive interpolation algorithm: The 15-minute plan curve P issued by the scheduler... plan (t) is converted into a 1-minute control instruction P. control (t).

[0143] Linear interpolation formula (original):

[0144] ;

[0145] in, This indicates that at the interpolation point t k Located at the original point t j Between; P plan (t) j ) indicates scheduling at t j Timely planned output; P control (t) k ) indicates that interpolation generates t k Time-based control commands.

[0146] In the description of this invention, the collaborative interpolation control module 3 includes: an output acquisition module, a regression analysis module, and a control execution module.

[0147] The output acquisition module is used to set interval time points and periodically acquire the scheduling plan curve according to the interval time points to represent the planned output of a hydropower unit; for example, it can acquire a point every 15 minutes from the scheduling system to represent the planned output of a unit, for a total of 96 points per day.

[0148] The regression analysis module estimates the derivative based on the historical operating data of the hydropower unit and substitutes it into Hermite interpolation to generate control points every minute. If the difference between the interpolation point and the actual operating data is greater than the preset threshold, an alarm is triggered and a switch is made, switching to the backup interpolation method.

[0149] Specifically, based on historical operating data, M is estimated using the finite difference method or regression model. j This includes the following steps:

[0150] 1. Obtain planned curve data: The scheduling system provides 96 output points Q per day. j 15 minutes apart.

[0151] 2. Estimate the derivative M j The midpoint is calculated using the central difference formula:

[0152] ;

[0153] Boundary points are determined using forward or backward differences.

[0154] 3. Import interpolation formula: Import the estimated M j Substitute the Hermite interpolation values ​​to generate control points per minute.

[0155] 4. Error verification: If the interpolation point deviates too much from the actual SCADA data, switch to the backup interpolation method (such as linear).

[0156] This can be summarized as: finite difference method for derivative estimation, importing Hermite interpolation, generating minute control points for real-time verification.

[0157] The control execution module is used to control the execution of data at two adjacent raw time intervals (P... j P j Between +1), control points {P} will be generated according to minute interpolation. k The data is then transmitted to the local control unit.

[0158] Furthermore, regarding the dynamic error threshold, that is: for each calculated P... k Each value must be compared with the real-time SCADA feedback value; if the difference exceeds the set threshold, an alarm will be triggered and the algorithm will switch to the backup interpolation method.

[0159] ;

[0160] in, ; These are Hermite basis functions; The endpoint derivative is obtained by fitting historical data.

[0161] Real-time verification mechanism: Define dynamic error threshold ;

[0162] like Trigger a Level 3 alarm and switch to the backup interpolation strategy.

[0163] V-curve recognition module 4 is used to construct spatiotemporal composite V-curve recognition rules and load change inertial transition strategies to avoid equipment losses and oscillation risks caused by the start-up and shutdown of hydropower units.

[0164] Specifically, the spatiotemporal composite judgment rule is a crucial logic in this system for identifying risk periods during the "V-shaped curve." It ensures that protective strategies such as "lock-in shutdown" and "smooth adjustment" are implemented in sensitive sections where the unit may face oscillations, accidental shutdowns, or impact loads. For example, a unit currently operating at 80MW load, with a real-time load check showing planned output of 80MW, 65MW, and 90MW, takes only 2 minutes. At this point: the system judges that it is within a shutdown window; and the load has suddenly changed; it is determined to be in a "V-shaped region"; the shutdown command is temporarily suspended; the load is adjusted using an inertial function; and the shutdown timing is reassessed after the situation stabilizes.

[0165] Time window Dynamic adjustment: ,in This refers to the average start-up time of the unit;

[0166] Output demand mutation detection: If it exists ,satisfy If so, it is marked as a V-shaped time period, and shutdown is prohibited; Calculated dynamically based on unit capacity.

[0167] The intelligent diagnostic module 5 is used to build a multi-level fault diagnosis framework to achieve fault location and self-healing.

[0168] Specifically, multi-level fault classification:

[0169] Primary classification: Mechanical failure (code 1xxx), Electrical failure (code 2xxx), Communication failure (code 3xxx);

[0170] Secondary classification: For example, mechanical failures are classified as bearing overheating (code 1100), insufficient lubrication (code 1200), etc.

[0171] The above situation occurs during automatic start-up and shutdown when a fault alarm function is triggered. Moreover, there are multiple fault classifications. In order to improve the accuracy and operability of fault diagnosis and avoid the problem of "one alarm message, unclear location" in the past, for example: Unit 4 alarms, abnormal vibration; the intelligent fault diagnosis module reports: Unit 3 alarms, mechanical fault, bearing overheating. Suggestion: check the cooling system and run at reduced load for 30 minutes.

[0172] Self-healing strategy recommendation: For example, "Unit 3 failed to start, fault code 1100 (bearing overheating), suggestion: 1. Check the cooling system; 2. Reduce the load and run for 30 minutes."

[0173] Risk scoring module 6, also known as FMEA scoring module, is used to calculate risk priority numbers based on failure mode analysis of key components of hydropower units and to mark them as fault risk scores integrated into the dynamic priority weight calculation model.

[0174] Specifically, the intelligent diagnostic module 5 is responsible for identifying and classifying fault types, such as bearing overheating and communication interruption, and providing the specific location of the fault and suggested handling strategies. On this basis, the risk scoring module 6 quantifies the risk level of various faults, calculates the risk priority number (RPN) to form a fault risk score γ value, and incorporates it into the priority weight calculation model to affect the start-up and shutdown strategy.

[0175] In the weighted calculation model, the original "fault risk score" item is further refined:

[0176] ;

[0177] Wherein, RPN (Risk Priority Number) is the risk priority number calculated by FMEA, which is normalized and used as the γ weight input; RPN max This represents the maximum risk priority number.

[0178] In the description of this invention, the risk scoring module 6 includes: a failure mode identification module, an impact consequence assessment module, an occurrence probability assessment module, a detection difficulty assessment module, a risk priority number calculation module, a normalization aggregation module, a priority weight merging module, and a feedback closed-loop optimization module.

[0179] The failure mode identification module is used to identify all failure modes of hydroelectric generator components (such as turbine bearings, guide vane mechanisms, generator stator windings, LCU communication modules, etc.), such as bearing overheating, guide vane jamming, stator winding short circuit, and communication interruption.

[0180] The impact and consequences assessment module is used to assess the impact of each failure mode on the operation of the hydropower unit.

[0181] Specifically, the impact of each failure mode on unit operation is assessed, such as the risk of shutdown delay / false shutdown, load fluctuation, and equipment damage, and scored on a "severity (S)" scale, ranging from 1 to 10, with 10 being the most severe.

[0182] The probability assessment module is used to score the frequency of occurrence of each failure mode based on historical operating data and hydropower unit failure statistics.

[0183] Specifically, based on historical operating data and failure statistics, the frequency of occurrence of each failure mode is scored, ranging from 1 to 10, with 10 indicating the most frequent occurrence.

[0184] The detection difficulty assessment module is used to score the difficulty of detecting each failure mode based on existing diagnostic methods.

[0185] Specifically, based on existing diagnostic methods, a score is given for whether the failure can be detected in a timely manner, ranging from 1 to 10, where 1 indicates easy detection and 10 indicates difficult detection.

[0186] The risk priority number calculation module is used to calculate the risk priority number for each failure mode.

[0187] Specifically, the formula for calculating the risk priority number is as follows:

[0188] ;

[0189] In the formula, S represents severity; O represents occurrence (frequency of occurrence); and D represents detectability.

[0190] The normalization aggregation module is used to set the maximum value of the risk priority number, normalize the risk priority number of each failure mode with the maximum value of the risk priority number, and perform a weighted average of the normalized risk priority numbers of all failure modes of the same hydropower unit to obtain a comprehensive failure risk score.

[0191] Specifically, set a reference maximum RPN value, such as the historical maximum or theoretical maximum of 1000, and normalize the RPN for each failure mode:

[0192] ;

[0193] The comprehensive failure risk score of a unit is obtained by summing or weighting the normalized RPNs of all failure modes for the same unit.

[0194] The priority weighting module is used to replace the corresponding item in the weighting calculation model with the comprehensive fault risk score, and to dynamically optimize the output results of the priority weighting of hydropower units.

[0195] Specifically, in the dynamic priority scoring function, the original γ term is replaced with the normalized fault risk score:

[0196] ;

[0197] The higher the RPN (the greater the risk of failure), the lower the priority score of the unit, and the automatic shutdown / switching sequence will be advanced or delayed accordingly.

[0198] The feedback closed-loop optimization module is used to update the maximum value of the risk priority number based on the trend of risk priority number changes, and optimize the evaluation criteria and weight parameters through a self-learning mechanism.

[0199] Specifically, based on the latest fault records, the S, O, and D scores are reassessed periodically, and the maximum RPN value is updated. In the "closed-loop optimization" step, the RPN change trend is used as part of the penalty / reward function to optimize the FMEA evaluation criteria and weight parameters.

[0200] To enhance the dynamic adaptability of the FMEA module, the system introduces the RPN change trend as part of the penalty / reward function and optimizes the FMEA scoring criteria and weight parameters through a self-learning mechanism.

[0201] In the description of this invention, the maximum value of the risk priority number is updated based on the trend of risk priority number changes, and the evaluation criteria and weight parameters are optimized through a self-learning mechanism, including:

[0202] Step S61: According to the control cycle, periodically record the risk priority number of each failure mode of each hydropower unit, and construct a time series of risk priority numbers. .

[0203] Step S62: Calculate the rate of change of risk priority number for any failure mode over two consecutive control cycles, compare it with the preset rate of change threshold, and determine whether it is in an increasing, normal and stable, or decreasing trend.

[0204] Specifically, calculate the RPN change rate for a certain failure mode over two consecutive periods:

[0205] ;

[0206] The threshold is set to determine whether the increase is abnormal, the condition is stable, or the decrease is significant.

[0207] Step S63: If the rate of change of risk priority number is increasing, it is determined that the diagnostic function has not controlled the risk in time, and the severity or frequency score of the current failure mode is increased; if the rate of change of risk priority number is decreasing, the severity or frequency score of the current failure mode is decreased.

[0208] Specifically, if ΔRPN increases significantly, it is determined that the system diagnosis failed to control the risk in a timely manner, and the severity (S) or frequency (O) score of the current failure mode is increased.

[0209] If ΔRPN continues to decline, it is considered that the risk control is effective, and the corresponding score is reduced.

[0210] Step S64: Evaluate the impact of the current failure mode parameters on the overall risk priority number, set minimizing the error as the objective function, and optimize the parameter combination to maintain the matching effect between the overall priority model score and the actual operating effect.

[0211] Specifically, assess the impact of the current FMEA three parameters (S, O, D) on the overall RPN;

[0212] Introduce gradient descent or genetic algorithm to optimize parameter combination so that the overall priority model score matches the actual running results (failure rate, number of downtimes);

[0213] Minimize the error using the following objective function:

[0214] .

[0215] Step S65: After the fault risk score for each control cycle is completed, the optimization results are written into the database.

[0216] Specifically, after each round of FMEA score update, the optimization results are written to the database; if there is no improvement after several rounds of optimization, the algorithm strategy can be switched (such as switching to particle swarm optimization) or some parameters can be reset; the optimized score is directly used for the next round of dynamic priority scoring model γ item.

[0217] Example tables are shown in Table 1.

[0218] Table 1: Scoring Examples

[0219]

[0220] Please see Figure 2 This paper presents an intelligent start-up and shutdown method for hydropower units based on multi-dimensional dynamic optimization and high-precision collaborative control. The method includes:

[0221] S1. Monitor the status of hydropower units in real time and collect operating data.

[0222] S2. Based on head, vibration zone, fault signals and historical operating data, priority weights are dynamically generated to achieve global optimization of the hydropower unit startup sequence.

[0223] S3. Convert the scheduling plan curve into control commands and introduce a dynamic error threshold verification mechanism to trigger multi-level alarms and optimize response speed.

[0224] S4. Construct spatiotemporal composite V-shaped curve identification rules and load change inertial transition strategies to avoid equipment losses and oscillation risks caused by the start-up and shutdown of hydropower units.

[0225] S5. Construct a multi-level fault diagnosis framework to achieve fault location and self-healing.

[0226] S6. Based on the failure mode analysis of key components of hydropower units, calculate the risk priority number and mark it as a fault risk score to be integrated into the dynamic priority weight calculation model.

[0227] The following detailed description of the invention is provided in conjunction with specific embodiments.

[0228] like Figure 3 As shown, the system of this invention adopts a distributed hardware architecture, deploying a real-time interactive module in the central control center or plant server, and connecting the data link through the SCADA system. The hardware layer integrates a high-precision sensor network and communication interface to achieve synchronous acquisition of multi-source data such as water head, vibration, and temperature, providing support for dynamic decision-making in the software layer. The core software function focuses on multi-unit collaborative control, supporting flexible switching between semi-automatic and fully automatic modes. During switching, safety checks such as technical water supply redundancy and lockout status are required to ensure the reliability of mode conversion.

[0229] Data stream processing is divided into two-way channels: the uplink receives the 15-minute scheduling plan curve, and generates second-level control commands through the interpolation algorithm verification module. If the deviation between the interpolation point and the measured data exceeds the threshold, multi-level alarms are triggered; the downlink sends the optimized start and stop commands to the unit's local control unit in real time, forming a "plan-control-execution" closed loop.

[0230] The start-up and shutdown strategy relies on a dynamic priority mechanism. Initial priorities are set based on unit capacity and historical failure rates. During operation, weights are dynamically adjusted based on real-time head adaptability, vibration zone offset (updated via sliding window self-learning), and fault signals. Start-up and shutdown sequence optimization employs the Hungarian algorithm, constructing a cost matrix of unit-time position. For example, units with high vibration risk are assigned to high-efficiency operating zones. The globally optimal start-up and shutdown sequence is output and iterated periodically. The control logic includes dual predictions: when a sharp drop in plant-wide power generation demand is predicted, the shutdown process is triggered in advance; when identifying V-shaped curve characteristic segments, an anti-false shutdown strategy is implemented—through ingot locking self-checks, minimum output maintenance, and AGC collaborative load sharing—avoiding unnecessary shutdowns during off-peak periods, as is common in traditional systems.

[0231] In special scenario handling, an inertial transition strategy is adopted for sudden load changes: if a rebound in load demand is detected within the shutdown countdown, the shutdown process is immediately terminated and the system switches to power-up mode, while the interpolation command slope is adjusted to quickly restore output. V-curve recognition is based on spatiotemporal composite rules, combining planned curve shutdown conditions with real-time vibration over-limit markers to dynamically lock the high-risk group status.

[0232] The human-machine interface integrates multi-dimensional operation feedback, and function buttons support one-click intervention of start-stop sequences. The alarm window pushes equipment-level diagnostic results in real time, such as "cooling main pump failed, switched to standby." The log system fully records priority adjustment events, vibration zone boundary update operations, and self-healing command execution times, providing data traceability for the risk scoring module's self-learning optimization. All events are tagged with operating conditions and stored in the database to support subsequent parameter tuning and strategy upgrades.

[0233] The system functions and main process steps of this invention can be summarized as follows.

[0234] I. System Initialization

[0235] Load unit parameters (rated capacity, vibration zone range, head efficiency curve);

[0236] Initialize priority weights and set interpolation algorithm parameters, such as the Hermite interpolation smoothing factor;

[0237] Connect to the SCADA system and calibrate the real-time data acquisition channel.

[0238] II. Real-time control loop

[0239] The real-time control loop is the main execution thread connecting the various modules of the "scheduling algorithm execution feedback," and it is the core of the system's continuous operation every second and every minute. It collects data once per second and updates priorities, interpolation, and execution control every minute.

[0240] Data Acquisition: Acquired every second Data can be collected and communicated with computer SCADA systems via UDP or IEC104.

[0241] Dynamic priority update: calculated by weight model Generate a priority queue;

[0242] Construct a dynamic priority scoring mechanism that can reflect the overall operating status, historical performance, vibration risk, and fault level of the unit, so as to reasonably determine "who starts first and who stops later".

[0243] The scoring function is as follows:

[0244] ;

[0245] in: H is the average output integral over a past period; i The current water head; The deviation of the vibration zone (0 for safety, 1 for criticality); F i A fault risk score is assigned; α, β, γ, and δ are weighting coefficients that can be learned to optimize the initial settings and reinforcement learning.

[0246] Interpolation and verification: Execute Hermite interpolation to generate a 1-minute command, and send it to LCU after successful verification;

[0247] Special scenario handling: Detect V-shaped time periods and immediately lock the shutdown command;

[0248] When the load changes abruptly, it is smoothly adjusted to the new target value based on the inertia coefficient. :

[0249] .

[0250] Troubleshooting: If startup fails, switch to standby unit according to priority and push diagnostic report to maintenance terminal.

[0251] III. Logs and Optimization

[0252] Record all operation events, fault codes, and control effects;

[0253] The adaptive adjustment coefficients α, β, γ, and δ are periodically optimized through reinforcement learning.

[0254] This system supports the self-learning evolution of the adjustment coefficient, which is achieved using reinforcement learning principles.

[0255] 1) Use historical output deviation, oscillation amplitude, and failure frequency as reward functions;

[0256] 2) Obtain performance scores by periodically testing different weight combinations;

[0257] 3) Employ strategy optimization algorithms (such as gradient descent or genetic search) to retain the optimal parameter set;

[0258] 4) Replace the current control strategy in the next cycle to form an adaptive scheduling capability.

[0259] In summary, by leveraging the technical solutions described above, the regulation value of hydropower in the new power system is significantly enhanced through the deep integration of dynamic priority optimization, high-precision command coordination, and intelligent diagnostic mechanisms. This overcomes the limitations of traditional static priority, dynamically adjusting the start-up and shutdown sequence based on real-time head changes, vibration zone offsets, and unit health status, thus avoiding equipment damage caused by inadvertently entering the vibration zone under high load conditions. For coarse-grained scheduling curves, a piecewise cubic Hermite interpolation algorithm is used to generate second-level control commands, coupled with the coordinated timing of the governor, excitation system, and synchronizing device. Dynamic error verification is used to quickly correct output during sudden load changes. Deviation is addressed to achieve precise matching with grid frequency fluctuations; a spatiotemporal composite V-shaped curve recognition mechanism is introduced to predict load troughs caused by sharp drops in renewable energy output, maintaining minimum technical output of units rather than frequent start-ups and shutdowns, effectively reducing mechanical stress on equipment; a multi-level fault diagnosis framework locates general alarms to specific equipment components, and combined with a self-healing recommendation mechanism, quickly switches to backup equipment or generates maintenance work orders, significantly reducing the risk of unplanned downtime; finally, through multi-module collaborative closed-loop, the success rate of unit start-up and shutdown, regulation response speed, and fault location accuracy are comprehensively improved under complex operating conditions, providing safe, efficient, and flexible regulation support for high-proportion renewable energy grids.

[0260] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

Claims

1. A smart start-up and shutdown system for hydropower units based on multi-dimensional dynamic optimization and high-precision collaborative control, characterized in that, include: The data acquisition module is used to monitor the status of the hydropower unit and collect operating data in real time; The dynamic priority management module is used to dynamically generate priority weights based on head, vibration zone, fault signals and historical operating data, so as to achieve global optimization of the hydropower unit startup sequence. The collaborative interpolation control module is used to convert scheduling plan curves into control commands and introduces a dynamic error threshold verification mechanism to trigger multi-level alarms and optimize response speed. The V-curve recognition module is used to construct spatiotemporal composite V-curve recognition rules and load change inertial transition strategies to avoid equipment losses and oscillation risks caused by the start-up and shutdown of hydropower units. The intelligent diagnostic module is used to build a multi-level fault diagnosis framework to achieve fault location and self-healing. The risk scoring module is used to calculate risk priority numbers based on failure mode analysis of key components of hydropower units and to mark them as fault risk scores integrated into the dynamic priority weight calculation model. The collaborative interpolation control module includes: The power output acquisition module is used to set interval time points and periodically acquire the scheduling plan curve according to the interval time points, which represents the planned power output of a hydropower unit; The regression analysis module estimates the derivative based on the historical operating data of the hydropower unit and substitutes it into the Hermite interpolation to generate control points every minute. If the difference between the interpolation point and the actual operating data is greater than the preset threshold, an alarm is triggered and the system switches to the backup interpolation method. The control execution module is used to transmit control points generated by minute interpolation to the local control unit between two adjacent original time intervals.

2. The intelligent start-up and shutdown system for hydropower units based on multi-dimensional dynamic optimization and high-precision collaborative control according to claim 1, characterized in that, The dynamic priority management module includes: The priority weight calculation module is used to integrate water head, vibration zone, fault signals and historical operating data to build a weight calculation model and calculate the priority weight at the current moment. The vibration zone self-learning module is used to adaptively update the vibration zone boundary based on historical operating data, and feed the updated vibration zone range back to the start-up and shutdown control logic to dynamically optimize the crossing strategy. The start-stop sequence optimization module is used to optimize the start-stop sequence of hydropower units using the Hungarian algorithm to maintain the overall operating efficiency of the hydropower units at the target level.

3. The intelligent start-up and shutdown system for hydropower units based on multi-dimensional dynamic optimization and high-precision collaborative control according to claim 2, characterized in that, The vibration zone self-learning module includes: The vibration acquisition module is used to collect operating data of the hydropower unit, including speed, power, vibration value, speed changes during start-up and shutdown, and abnormal vibration characteristics during start-up and shutdown. The initial determination module is used to set the initial vibration zone range, which serves as the initial input for the vibration zone model. The annotation and training module is used to perform offline analysis of historical operating data, extract the rotational speed range of abnormal vibration points, statistically analyze the high vibration range under similar working conditions, and use machine learning to build a model to generate a vibration zone model. It also extracts and outputs the target vibration frequency band through cluster analysis. The online self-learning module is used to introduce a sliding window algorithm to update the vibration zone range in real time, feed the updated vibration zone range back to the start-up and shutdown control logic, and perform safety fault tolerance verification.

4. The intelligent start-up and shutdown system for hydropower units based on multi-dimensional dynamic optimization and high-precision collaborative control according to claim 3, characterized in that, The introduced sliding window algorithm updates the vibration zone range in real time, feeds the updated vibration zone range back to the start-up and shutdown control logic, and performs safety fault tolerance verification, including: Set the parameters and data fields for the sliding window; Acquire real-time hydropower unit operation data, extract vibration data, load and speed, add new data points to a sliding window, and remove old data that exceeds the constant window length; Calculate the mean and variance of the vibration value corresponding to each load segment within the sliding window, detect whether there is an abnormal vibration interval exceeding the preset threshold and preset duration, if so, mark it as an abnormal vibration segment, and dynamically adjust it using the exponential weighted moving average method; The number of abnormal markings for load segments is counted. If the number of consecutive markings of a load segment as an abnormal vibration segment exceeds a preset threshold, the abnormal vibration segment is written into the vibration zone model cache. The new vibration zone model is compared with the original vibration zone model. If the model deviation exceeds the set threshold, the replacement is temporarily suspended. If the model deviation is within the preset range, passes multiple consecutive verifications, and passes the working condition consistency verification, the new vibration zone model replaces the original vibration zone model. The updated vibration zone model's vibration zone range is fed back into the start-up and shutdown control logic to dynamically optimize the crossing strategy and modify the vibration risk item in the start-up and shutdown priority scoring.

5. The intelligent start-up and shutdown system for hydropower units based on multi-dimensional dynamic optimization and high-precision collaborative control according to claim 4, characterized in that, The dynamic adjustment using the exponentially weighted moving average method includes: The vibration value of the vibration zone model at the current moment is collected, combined with the smoothed output value of the previous moment, and the smoothed output value at the current moment is calculated using the exponentially weighted average working hours. The smoothed output value is used as the basis for determining the new vibration zone, and the model boundary of the vibration zone model is updated. If the real-time vibration value is higher than the updated model boundary for more than a preset time, the control strategy is adjusted.

6. The intelligent start-up and shutdown system for hydropower units based on multi-dimensional dynamic optimization and high-precision collaborative control according to claim 2, characterized in that, The start / stop sequence optimization module includes: The initialization module is used to load the basic parameters of the hydropower units and the time window and time slot division of the current scheduling plan. At the beginning of each control cycle, it obtains the real-time priority weight of each hydropower unit. The matrix construction module is used to set the set of hydropower units and the set of sequence positions to be assigned, forming sequence pairs of units and sequence positions. For each sequence pair, an allocation cost is set to form a cost matrix. The optimal matching module is used to take the cost matrix as input, call the Hungarian algorithm, and obtain a set of one-to-one matching that minimizes the total cost. The generation and distribution module is used to arrange the units corresponding to each sequence position in sequence according to the matching result of minimizing the total cost, obtain the optimal start-stop sequence of the current control cycle, and convert the current sequence into start-stop commands, which are then distributed to the local control units of each hydropower unit for execution. The iterative feedback module is used to repeatedly perform sequence optimization operations within a preset iteration cycle, dynamically respond to changes in head, vibration zone shifts and fault conditions, and record the running effect and deviation of each match. The adaptive adjustment coefficient in the weight calculation model is optimized synchronously through logs and reinforcement learning.

7. The intelligent start-up and shutdown system for hydropower units based on multi-dimensional dynamic optimization and high-precision collaborative control according to claim 1, characterized in that, The risk scoring module includes: Failure Mode Identification (FMO) module is used to organize all failure modes of hydropower unit components; The impact and consequences assessment module is used to assess the impact of each failure mode on the operation of the hydropower unit; The probability assessment module is used to score the frequency of occurrence of each failure mode based on historical operating data and hydropower unit failure statistics. The detection difficulty assessment module is used to score the difficulty of detecting each failure mode based on existing diagnostic methods; The risk priority number calculation module is used to calculate the risk priority number for each failure mode. The normalization aggregation module is used to set the maximum value of the risk priority number, normalize the risk priority number of each failure mode with the maximum value of the risk priority number, and perform a weighted average of the normalized risk priority numbers of all failure modes of the same hydropower unit to obtain a comprehensive failure risk score. The priority weighting module is used to replace the corresponding item in the weighting calculation model with the comprehensive fault risk score, and to dynamically optimize the output results of the priority weighting of hydropower units. The feedback closed-loop optimization module is used to update the maximum value of the risk priority number based on the trend of risk priority number changes, and optimize the evaluation criteria and weight parameters through a self-learning mechanism.

8. The intelligent start-up and shutdown system for hydropower units based on multi-dimensional dynamic optimization and high-precision collaborative control according to claim 7, characterized in that, The process of updating the maximum value of the risk priority number based on its changing trend, and optimizing the evaluation criteria and weight parameters through a self-learning mechanism, includes: According to the control cycle, the risk priority number of each failure mode of each hydropower unit is recorded periodically to construct a time series of risk priority numbers; Calculate the rate of change of risk priority number for any failure mode over two consecutive control cycles, compare it with a preset rate of change threshold, and determine whether it is increasing, stable, or decreasing. If the rate of change of risk priority number is increasing, it is determined that the diagnostic function has not controlled the risk in a timely manner, and the severity or frequency score of the current failure mode is increased; if the rate of change of risk priority number is decreasing, the severity or frequency score of the current failure mode is decreased. Assess the impact of current failure mode parameters on the overall risk priority number, set minimizing the error as the objective function, and optimize the parameter combination to maintain the matching effect between the overall priority model score and the actual operating effect; After the fault risk score is completed for each control cycle, the optimization results are written into the database.

9. A method for intelligent start-up and shutdown of hydropower units based on multi-dimensional dynamic optimization and high-precision collaborative control, employing the intelligent start-up and shutdown system for hydropower units based on multi-dimensional dynamic optimization and high-precision collaborative control as described in any one of claims 1-8, characterized in that, The method includes: S1. Real-time monitoring of the status of hydropower units and collection of operating data; S2. Based on head, vibration zone, fault signals and historical operating data, priority weights are dynamically generated to achieve global optimization of the hydropower unit startup sequence; S3. Convert the scheduling plan curve into control commands and introduce a dynamic error threshold verification mechanism to trigger multi-level alarms and optimize response speed; S4. Construct spatiotemporal composite V-shaped curve identification rules and load change inertial transition strategies to avoid equipment losses and oscillation risks caused by the start-up and shutdown of hydropower units; S5. Construct a multi-level fault diagnosis framework to achieve fault location and self-healing; S6. Based on the failure mode analysis of key components of hydropower units, calculate the risk priority number and mark it as a fault risk score to be integrated into the dynamic priority weight calculation model.

Images