Offshore wind power operation and maintenance method considering uncertainty and opportunity collaborative maintenance

Abstract

The invention discloses an offshore wind power operation and maintenance method considering uncertainty and opportunity collaborative maintenance. The method is used for solving the technical problems that an offshore wind plant is high in operation and maintenance cost and large in uncertainty. The method comprises the following steps: firstly, acquiring wind speed, wave height, electricity price, wind speed and point prediction data thereof, and a prediction value of residual life of a wind turbine generator component; generating a group uncertain scene as the input of a subsequent optimization model through a space-time probability model; constructing a two-stage random mixed integer programming model as an optimization model, and setting decision variables and constraint conditions of the model; and finally, solving the two-stage random mixed integer programming model every day by adopting a rolling time domain optimization algorithm, and outputting a maintenance scheduling plan. The embodiment proves that compared with an existing advanced scheduling strategy, the offshore wind power operation and maintenance economy can be further improved, and the availability and robustness of the scheduling strategy are enhanced.

Description

Technical Field

[0001] This invention relates to the field of intelligent operation and maintenance technology for offshore wind power, and in particular to an offshore wind power operation and maintenance method that considers both uncertainty and opportunity in collaborative maintenance. Background Technology

[0002] Against the backdrop of driving global energy structure transformation and mitigating climate change, offshore wind power, due to its clean and efficient characteristics, has become a key direction for renewable energy development. However, compared to onshore wind power, the high operation and maintenance costs of offshore wind power severely restrict its economic benefits and sustainable development capabilities. This is mainly due to its unique operating environment and technical challenges, such as the remote location of offshore wind farms, harsh weather conditions, difficulty in scheduling operation and maintenance resources, accelerated degradation of turbine components, and significantly higher downtime losses compared to onshore wind power. Existing maintenance strategies, such as fault maintenance, preventive maintenance, and opportunistic maintenance, appear inefficient and expensive in this context, and are unable to cope with the complex operation and maintenance task optimization and scheduling problems brought about by complex operating environments, fluctuating electricity market prices, difficulties in scheduling operation and maintenance resources, and accelerated equipment degradation. Therefore, there is an urgent need for an intelligent operation and maintenance scheduling strategy that can fully utilize the operation and maintenance environment, economic opportunities, and operation and maintenance resource opportunities, and possess high adaptability to cope with uncertainties in the environment, economy, and turbine degradation. Summary of the Invention

[0003] The purpose of this invention is to overcome the shortcomings of the existing technology and provide an offshore wind power operation and maintenance method that considers both uncertainty and opportunity in its maintenance, which has stronger robustness, economy and reliability compared to the existing technology.

[0004] The above objectives are achieved through the following technical solutions:

[0005] The present invention provides an offshore wind power operation and maintenance method that considers both uncertainty and opportunity in its collaborative maintenance, comprising the following steps:

[0006] S1: Collects point prediction data on wind speed, wave height, electricity price, wind speed, wave height, electricity price, and predicted remaining lifespan of wind turbine components. ;

[0007] S2: Generated through a spatiotemporal probability model A group of uncertain scenarios, among which Indicates the number of scenes:

[0008] By combining wind speed and wave height uncertainties with weather maintenance thresholds, the weather opportunities for operation and maintenance are assessed; by combining wind speed-power generation relationship with electricity price uncertainty, the low power generation loss opportunities for operation and maintenance are assessed; based on the remaining lifespan of unit components, the maintenance method of preventive maintenance or fault maintenance is evaluated by the subsequent optimization model; the above results will be used as inputs for the subsequent optimization model.

[0009] S3: Construct a two-stage stochastic mixed-integer programming model as the optimization model, and set the decision variables and constraints of the model;

[0010] S4: Use a rolling horizon optimization algorithm to solve the two-stage stochastic mixed-integer programming model daily and output the maintenance scheduling plan.

[0011] Further, the specific method of step S2 comprises:

[0012] S21. The wind speed, wave height, and electricity price uncertainty scenario generation is based on: wherein, is the predicted variable, representing the value of the wind speed / wave height / electricity price at time ; is the prior wind speed / wave height / electricity price value; is the residual value calculated based on the zero-mean Gaussian process regression model; is white noise satisfying independent and identical distribution, , is the noise variance, and the joint distribution of the wind speed, wave height, and electricity price uncertainty scenario is obtained through training of the Gaussian process regression model, and a plurality of uncertainty scenarios are obtained through a plurality of random samplings;

[0013] S22. The wind turbine remaining life prediction is based on:

[0014] It is assumed that the remaining life of each unit obeys a Weibull distribution: wherein, is a random variable representing the predicted remaining life of the unit ; is a scale parameter of the Weibull distribution predicted by the monitoring system for the unit component life; is a shape parameter of the Weibull distribution; a plurality of remaining life values of the unit component under a plurality of uncertainty scenarios are generated through random sampling of , and the maintenance mode of the unit under each scenario is determined to be preventive maintenance or failure maintenance.

[0015] Further, the step of predicting based on the zero-mean Gaussian process regression model comprises the following steps:

[0016] The predicted value and its distribution at the future time are modeled through the following steps, and the model implicitly includes the spatiotemporal correlation of the residual process :

[0017] S211. Construct a covariance matrix , size of , is the current time point, the element in is: where, is the indicator function, which is 1 when , otherwise 0; is the covariance function, is the noise variance;

[0018] S212. Covariance function selection:

[0019] Since the wind speed, wave height and electricity price have strong time correlation, the covariance function is chosen as the square exponential kernel: where, is the signal variance, is the length scale, is the exponential function;

[0020] S213. Model training, i.e. hyperparameter estimation:

[0021] The hyperparameters are estimated by maximizing the marginal likelihood: where, is the historical observation; is the corresponding original point prediction value, is the logarithm of the marginal likelihood function, is the logarithm of the determinant of the covariance matrix, is the natural logarithm function, is the constant pi, and the superscript T is the matrix transpose,

[0022] S214. Prediction of the future step, i.e. the value at time , whose conditional distribution is Gaussian distribution, with mean:

[0023] and variance: where, is the prediction value of the original point prediction model at time ; is the vector whose elements are the historical time With the predicted time Covariance: ; The covariance matrix is ​​estimated based on historical data; The signal variance of the kernel function, i.e. ; For expectation calculation; For the predicted time The variance; Variance operator;

[0024] S215. Scene Generation: Generate a random trajectory for the next H steps. , i.e., scenario: In the formula, To predict the mean vector; To predict the covariance matrix, the size is , elements in for: In the formula, for Random trajectory at time, for Random trajectory at time, It is the covariance function;

[0025] Then, by analyzing random trajectories Several samplings were conducted to obtain several sets of random scenarios with uncertainties in wind speed, wave height, and electricity price. By combining the uncertain scenarios of wind speed and wave height with weather maintenance thresholds, the weather opportunities for operation and maintenance were assessed. By combining the wind speed-power generation relationship with the uncertain scenarios of electricity price, the low power generation loss opportunities for operation and maintenance were assessed.

[0026] Furthermore, the objective function of the two-stage stochastic mixed-integer programming model described in step S3 is: In the formula, For short-term gains, i.e., intraday profits. For the first Long-term profits for the day To maintain the cycle number of days, The number of uncertain scenarios; A collection of uncertain scenarios; A collection of units awaiting maintenance; For the scene Next Individual unit maintenance interruption indicator variable, when hour ; For the scene Next Remaining time for each unit to complete its task; Penalty factor related to maintenance interruption; Penalty costs for temporarily hiring maintenance personnel and for staff working overtime; This is an auxiliary variable that exceeds the budget for the number of maintenance personnel and the budget for overtime hours; The variable is 0-1, indicating whether short-term routine maintenance will be initiated for unit i at time t, where 1 indicates maintenance will be initiated and 0 indicates maintenance will not be initiated. This is a 0-1 variable, indicating whether long-term maintenance is performed on unit i on day d in scenario s. This is a 0-1 variable representing whether a vessel is leased for short-term maintenance; 0 indicates no leasing, and 1 indicates leasing. The variable is 0-1, indicating whether to rent a boat on day d in scenario s, where 0 means not to rent and 1 means to rent.

[0027] Short-term gains, i.e., intraday profits for: In the formula, Maintenance cycle duration; For the unit Running status This indicates that it is normal. Indicates a fault; For fault maintenance costs, To cover preventative maintenance costs, To determine whether maintenance began the previous day but was not completed, It means yes; For the unit Is it in STH? Maintenance is initiated at any time; 1 indicates maintenance is initiated, and 0 indicates maintenance is not performed. Regarding whether to charter a boat, Indicates leasing. Indicates that it is not rented; Daily rental fee for the vessel; To maintain the hourly wage of staff; Overtime pay for maintenance personnel (hourly rate); For the scene Are the maintenance personnel present? Constantly monitor the unit Performing maintenance (1 indicates maintenance is in progress, 0 indicates maintenance is not in progress); For the scene Overtime hours; For the scene Next moment Electricity price; For the scene Next moment unit Output power;

[0028] No. Long-term profits for: In the formula, Let d be the long-term electricity price forecast for day d under scenario s. Let i be the long-term output power of unit i on day d under scenario s. The variable is 0-1, representing the long-term operating status of unit i on day d under scenario s, where 1 indicates normal operation and 0 indicates shutdown for maintenance. This is a 0-1 variable, representing whether long-term maintenance is performed on unit i on day d under scenario s. 0 indicates no long-term maintenance, and 1 indicates long-term maintenance. Let d be the time required for long-term maintenance of unit i on day d under scenario s, in hours. Let d be the number of long-term overtime hours on day d under scenario s; For fault maintenance costs, To cover preventative maintenance costs, To determine whether maintenance began the previous day but was not completed, Yes, Indicates no.

[0029] Furthermore, the constraints of the stochastic mixed-integer optimization model described in step S3 include:

[0030] Task assignment uniqueness constraint: In the formula, represents each unit. Maintenance tasks must be scheduled in STH or LTH and only once;

[0031] Daily work time window constraints: The daily work time window constraint means that maintenance tasks can only be performed during the day. to It can be done between times, and cannot be earlier than time. No later than the time ;

[0032] Maintenance task continuity constraints: In the formula, To determine the time required to complete maintenance task i starting from time t in scenario s, the impact of weather accessibility is taken into account; the maintenance task continuity constraint means that after the task starts, the crew must remain in maintenance mode until it is completed or STH ends.

[0033] Maintenance interruptions and intertemporal constraints: Maintenance interruption and time-series constraints indicate the remaining time if a task is not completed in STH. It must continue at LTH and trigger the interruption penalty at this time. ;

[0034] Maintenance personnel allocation constraints: , The variable is 0-1, representing whether unit i is in maintenance state at time t under scenario s. 0 indicates not in maintenance state, and 1 indicates in maintenance state. The maintenance personnel allocation constraint indicates that if unit i is in maintenance state at time t, i.e. Then maintenance personnel need to be assigned. ;

[0035] Maintenance personnel number constraints: The maintenance personnel number constraint indicates that at any given time there can be a maximum of [number missing] maintenance personnel. The maintenance team's work exceeds the time limit, triggering a high-cost penalty; the excess portion is used... Indicates temporary employment;

[0036] Unit operating status update: In the formula, To indicate a specific moment during short-term maintenance, It is a 0-1 variable, representing the time of unit i at time... Whether maintenance is started: 0 indicates maintenance is not started, 1 indicates maintenance is started. This is a 0-1 variable, representing whether group i is available on day d in scenario s, where 0 indicates unavailable and 1 indicates available. This represents the total number of days in the long-term maintenance phase. Representing a scene Lower unit exist Is the time available? (0 indicates available, 0 indicates unavailable); the numerator ensures the fan is available again after maintenance; the denominator avoids division by zero. This is to avoid extremely small positive numbers with a denominator of 0;

[0037] Constraints are unavailable during maintenance: During maintenance, constraints cannot be used to indicate that the wind turbine cannot generate electricity while it is under maintenance.

[0038] Vessel scheduling and leasing constraints: In the formula, For a sufficiently large positive number, the vessel scheduling and leasing constraint means that if there is a planned maintenance task on a certain day, a vessel must be leased. (The law ensures the logic holds true).

[0039] Overtime constraints for maintenance personnel: In the formula, the overtime constraint for maintenance personnel means that the total working hours ≤ normal working hours (B×W) + overtime working hours. +Over-limit penalty On the first day of LTH, priority should be given to handling unfinished tasks, including those with remaining time. The daily overtime work hour limit is Hour;

[0040] Power output constraints: In the formula, For the scene Lower unit exist Electricity generation at any given time (MW); Rated capacity of the wind turbine (MW); The normalized power coefficient for wind speed; The long-term wind speed normalized power coefficient of unit i on day d under scenario s is between 0 and 1; during preventive maintenance, power generation is deducted proportionally. This is for preventative maintenance.

[0041] Grid absorption constraints: The grid absorption constraint means that the total output power is less than or equal to the grid's absorption capacity. Let be the grid absorption ratio at time t under scenario s, which is between 0 and 1.

[0042] Furthermore, the specific steps of the rolling time-domain algorithm described in step S4 are as follows:

[0043] S41. State initialization:

[0044] S411. Will Set to 1, A 0-1 variable, representing the unit The indicator shows whether the maintenance task has been completed; 1 indicates that it has not been completed, and 0 indicates that it has been completed.

[0045] S412. Will Set to 1, It is a 0-1 variable, indicating whether the unit has started maintenance, with 1 indicating that it has not started and 0 indicating that it has started; A 0-1 variable, representing the unit Whether it is available, 1 indicates available, 0 indicates unavailable;

[0046] S413. Set the current scroll wheel number Setting it to 0 indicates that the current scroll wheel number is the first scroll wheel;

[0047] S42. Enter the time-domain rolling loop;

[0048] S421. Set the short-term time window to... This indicates that the short-term time span of each round of rolling is 24 hours;

[0049] S422. Set the long-term time window to This indicates the long-term time span of each round of rolling. sky;

[0050] S43. Calculate state variables

[0051] Based on the already generated The scenarios include wind speed, wave height, electricity price, and remaining lifespan of generator components, combined with wind speed. High waves A safety threshold determines the duration of a maintenance task that can be performed, i.e., weather opportunity; based on wind speed. and electricity price Computerized maintenance of low power generation loss opportunities; determination of preventive or fail-safe maintenance based on the expected remaining lifespan of unit components;

[0052] S44. Solving a two-stage stochastic mixed integer programming model

[0053] The required data for the two-stage stochastic mixed integer programming model is input and the Gurobi solver is called to solve it, yielding the short-term scheduling variables. and long-term scheduling variables The values ​​are then combined to obtain the scheduling plan for this rolling solution;

[0054] S45. Status Update

[0055] Units that have completed maintenance tasks That Set to 0; units whose maintenance tasks have not been completed will have their maintenance tasks set to 0. Set the value to , Value represents the unit Remaining repair time when repairs are not completed in short-term scenarios; Set the value to 1. A 0-1 variable used to label the generator set. Whether it is "transferred" to the long-term maintenance phase, 1 indicates that it needs to be put into the long-term maintenance phase, and 0 indicates that it does not need to be put into the long-term maintenance phase;

[0056] S46. Rolling Iteration

[0057] Will Set as This indicates that the rolling iteration has entered the next round; and the life status of the wind turbine is updated based on the completion status of this maintenance task.

[0058] S47. Termination and Output

[0059] If all Then the algorithm will terminate rolling and all schedule results will be output as a maintenance plan.

[0060] The steps of the rolling time-domain optimization algorithm are as follows:

[0061] (1) Update the unit's maintenance status daily.

[0062] (2) Based on the collected data on wind speed, wave height, electricity price, and remaining lifespan of unit components, an uncertainty scenario generation model is constructed using a Gaussian process regression model and a Weibull distribution, and the model is generated. An uncertain scenario, then the computer group maintenance opportunity;

[0063] (3) Use the calculated data as input to the two-stage stochastic mixed integer programming model and call the Gurobi solver to solve the model;

[0064] (4) Integrate the solution results to construct a complete maintenance plan for the current iteration;

[0065] (5) Update the wind turbine status and model input data and use rolling iteration for the next round of optimization until all maintenance tasks are scheduled.

[0066] The advantages of this invention compared to the prior art are:

[0067] (1) Traditional offshore wind power maintenance methods mostly focus on cost savings brought about by resource sharing opportunities (such as simultaneous scheduling of equipment and personnel), while ignoring the superimposed effects of weather accessibility opportunities and low power generation loss opportunities (during periods of low electricity prices and wind speeds). This invention models and coordinates the three types of operation and maintenance opportunities in a unified manner, fully exploring their interactive potential. In this model, the unified consideration of the three types of opportunities can not only reduce conflicts and delays in maintenance tasks, but also improve the utilization rate of operation and maintenance resources and reduce revenue losses through task merging and collaborative scheduling, thereby achieving a systematic reduction in the total operation and maintenance cost.

[0068] (2) In actual offshore wind power operation and maintenance, key parameters such as wind speed, wave height, electricity price, and remaining lifespan of turbine components are subject to significant uncertainties. These uncertainties directly affect the availability and optimality of maintenance plans. Existing operation and maintenance technologies often consider deterministic scenarios, which can lead to maintenance plans deviating from real conditions, resulting in maintenance interruptions and delays, a surge in downtime losses, difficulty in ensuring the security of maintenance resources, and higher total operation and maintenance costs. This invention introduces a probabilistic prediction model into maintenance scheduling, generates random trajectories for multiple scenarios to characterize the uncertainties of the above parameters, and constructs a two-stage stochastic mixed integer optimization model that considers multiple sources of uncertainty. This allows the method to not only maintain a better maintenance plan when facing sudden weather changes or electricity price fluctuations, but also significantly reduce the high cost of operation and maintenance failures caused by relying on deterministic prediction results, greatly improving the robustness of the method under the influence of multiple uncertain factors.

[0069] (3) To adapt to the daily weather changes and information updates faced by wind farms, this invention designs a rolling time-domain scheduling algorithm to simulate the daily plan update situation at the industrial level. This algorithm updates weather and electricity price forecasts and unit operating status daily, generates a daily maintenance plan, and implements only the short-term plan portion. The long-term maintenance plan serves as a resource reserve reference for subsequent maintenance, thus achieving rolling time-domain optimization. Compared to existing static methods that formulate long-term plans all at once, rolling optimization can dynamically adjust the maintenance plan based on unexpected events, prediction errors, or resource changes during the operation and maintenance process, and achieve local optimal superposition through daily rolling. Therefore, this algorithm can adapt to engineering realities and avoid non-optimal or even high-cost infeasible decisions planned by static global optimization strategies due to changes in conditions, achieving the best balance between cost control and risk avoidance. Attached Figure Description

[0070] Figure 1 This is a schematic diagram of an offshore wind power operation and maintenance optimization method and system considering multi-source uncertainty and opportunity-based collaborative maintenance according to the present invention.

[0071] Figure 2This is a flowchart of a probabilistic prediction method, namely a Gaussian process regression model.

[0072] Figure 3 Flowchart of the rolling time-domain optimized scheduling algorithm;

[0073] Figure 4 Schematic diagrams are generated for uncertain scenarios, (a) showing an uncertain wind speed scenario, (b) showing an uncertain wave height scenario, and (c) showing an uncertain electricity price scenario.

[0074] Figure 5 This is a diagram illustrating the availability of the generating unit.

[0075] Figure 6 Gantt chart for unit maintenance plan;

[0076] Figure 7 A chart comparing the operation and maintenance costs and profits of different strategies. Figure 7 In the diagram, (a) shows the power generation profit graph for different strategies, and (b) shows the total operation and maintenance cost box diagram for different strategies. Detailed Implementation

[0077] Please refer to the specific steps for implementing this invention. Figure 1 For methods on generating uncertain scenarios, please refer to [link / reference]. Figure 2 The specific method is as follows:

[0078] S1: Collects point prediction data on wind speed, wave height, electricity price, wind speed, wave height, electricity price, and predicted remaining lifespan of wind turbine components. ;

[0079] S2: Generated through a spatiotemporal probability model A group of uncertain scenarios, among which Representing the number of scenes: Specific methods include:

[0080] S21. Scenario generation based on uncertain wind speed, wave height, and electricity price is based on: , In the formula, For predictor variables, wind speed / wave height / electricity price at time t = 0. The value; For a priori wind speed / wave height / electric value; These are the residual values ​​calculated based on a zero-mean Gaussian process regression model; To satisfy the requirement of independent and identically distributed white noise, , To represent the noise variance, the joint distribution of uncertain scenarios for wind speed, wave height, and electricity price is obtained by training a Gaussian process regression model, and several uncertain scenarios are obtained through several random samplings.

[0081] The steps for prediction based on the zero-mean Gaussian process regression model are as follows:

[0082] The following steps are used to predict values ​​for future times. The model is constructed to model the distribution of the residuals, which implicitly includes the residual process. Spatiotemporal correlation:

[0083] S211. Construct the covariance matrix , Size is , At the current time point, elements in for: In the formula, It is an indicator function, when The value is 1 if it is true, and 0 otherwise. It is the covariance function. It is the noise variance;

[0084] S212. Selection of covariance function:

[0085] Because wind speed, wave height, and electricity price have a strong time correlation, the covariance function... Selected as the squared exponent kernel: In the formula, For signal variance, For length scale, It is an exponential function;

[0086] S213. Model training, i.e., hyperparameter estimation:

[0087] Hyperparameters are estimated by maximizing marginal likelihood. : In the formula, These are historical observation values; The corresponding predicted value for the original point. Let be the logarithm of the marginal likelihood function. Let be the logarithm of the determinant of the covariance matrix. It is the natural logarithm function. Pi, with the superscript T representing matrix transpose.

[0088] S214. Predicting the Future Step, i.e., moment value Its conditional distribution is a Gaussian distribution with a mean of: The variance is: In the formula, For the original point prediction model in The predicted value at any given time; for A vector whose elements are historical moments. With the predicted time Covariance: ; The covariance matrix is ​​estimated based on historical data; The signal variance of the kernel function, i.e. ; For expectation calculation; For the predicted time The variance; Variance operator;

[0089] S215. Scene Generation: Generate a random trajectory for the next H steps. , i.e., scenario: In the formula, To predict the mean vector; To predict the covariance matrix, the size is , elements in for: In the formula, for Random trajectory at time, for Random trajectory at time, It is the covariance function;

[0090] Then, by analyzing random trajectories Several samplings were conducted to obtain several sets of random scenarios with uncertainties in wind speed, wave height, and electricity price. By combining the uncertain scenarios of wind speed and wave height with weather maintenance thresholds, the weather opportunities for operation and maintenance were assessed. By combining the wind speed-power generation relationship with the uncertain scenarios of electricity price, the low power generation loss opportunities for operation and maintenance were assessed.

[0091] S22. Wind turbine remaining life prediction is based on:

[0092] Assuming each unit The remaining lifetime follows a Weibull distribution: In the formula, To indicate the unit Random variables that predict remaining lifespan; To monitor the system and predict the lifespan of unit components, it serves as a scale parameter for the Weibull distribution; Let be the shape parameter of the Weibull distribution; by using Random sampling generates the remaining life values ​​of unit components under several uncertain scenarios, thereby determining whether to adopt preventive maintenance or fault maintenance for the unit under each scenario.

[0093] By combining wind speed and wave height uncertainties with weather maintenance thresholds, the weather opportunities for operation and maintenance are assessed; by combining wind speed-power generation relationship with electricity price uncertainty, the low power generation loss opportunities for operation and maintenance are assessed; based on the remaining lifespan of unit components, the maintenance method of preventive maintenance or fault maintenance is evaluated by the subsequent optimization model; the above results will be used as inputs for the subsequent optimization model.

[0094] S3: Construct a two-stage stochastic mixed-integer programming model as the optimization model, and set the decision variables and constraints of the model; the objective function of the two-stage stochastic mixed-integer programming model is: In the formula, For short-term gains, i.e., intraday profits. For the first Long-term profits for the day To maintain the cycle number of days, The number of uncertain scenarios; A collection of uncertain scenarios; A collection of units awaiting maintenance; For the scene Next Individual unit maintenance interruption indicator variable, when hour ; For the scene Next Remaining time for each unit to complete its task; Penalty factor related to maintenance interruption; Penalty costs for temporarily hiring maintenance personnel and for staff working overtime; This is an auxiliary variable that exceeds the budget for the number of maintenance personnel and the budget for overtime hours; The variable is 0-1, indicating whether short-term routine maintenance will be initiated for unit i at time t, where 1 indicates maintenance will be initiated and 0 indicates maintenance will not be initiated. This is a 0-1 variable, indicating whether long-term maintenance is performed on unit i on day d in scenario s. This is a 0-1 variable representing whether a vessel is leased for short-term maintenance; 0 indicates no leasing, and 1 indicates leasing. The variable is 0-1, indicating whether to rent a boat on day d in scenario s, where 0 means not to rent and 1 means to rent.

[0095] Short-term gains, i.e., intraday profits for: In the formula, Maintenance cycle duration; For the unit Running status This indicates that it is normal. Indicates a fault; For fault maintenance costs, To cover preventative maintenance costs, To determine whether maintenance began the previous day but was not completed, It means yes; For the unit Is it in STH? Maintenance is initiated at any time; 1 indicates maintenance is initiated, and 0 indicates maintenance is not performed. Regarding whether to charter a boat, Indicates leasing. Indicates that it is not rented; Daily rental fee for the vessel; To maintain the hourly wage of staff; Overtime pay for maintenance personnel (hourly rate); For the scene Are the maintenance personnel present? Constantly monitor the unit Performing maintenance (1 indicates maintenance is in progress, 0 indicates maintenance is not in progress); For the scene Overtime hours; For the scene Next moment Electricity price; For the scene Next moment unit Output power;

[0096] No. Long-term profits for: In the formula, Let d be the long-term electricity price forecast for day d under scenario s. Let i be the long-term output power of unit i on day d under scenario s. The variable is 0-1, representing the long-term operating status of unit i on day d under scenario s, where 1 indicates normal operation and 0 indicates shutdown for maintenance. This is a 0-1 variable, representing whether long-term maintenance is performed on unit i on day d under scenario s. 0 indicates no long-term maintenance, and 1 indicates long-term maintenance. Let d be the time required for long-term maintenance of unit i on day d under scenario s, in hours. Let d be the number of long-term overtime hours on day d under scenario s; For fault maintenance costs, To cover preventative maintenance costs, To determine whether maintenance began the previous day but was not completed, Yes, Indicates no.

[0097] The constraints of the stochastic mixed-integer optimization model described in step S3 include:

[0098] Task assignment uniqueness constraint: In the formula, represents each unit. Maintenance tasks must be scheduled in STH or LTH and only once;

[0099] Daily work time window constraints:

[0100] The daily work time window constraint means that maintenance tasks can only be performed during the day. to It can be done between times, and cannot be earlier than time. No later than the time ;

[0101] Maintenance task continuity constraints: In the formula, To determine the time required to complete maintenance task i starting from time t in scenario s, the impact of weather accessibility is taken into account; the maintenance task continuity constraint means that after the task starts, the crew must remain in maintenance mode until it is completed or STH ends.

[0102] Maintenance interruptions and intertemporal constraints:

[0103] Maintenance interruption and time-series constraints indicate the remaining time if a task is not completed in STH. It must continue at LTH and trigger the interruption penalty at this time. ;

[0104] Maintenance personnel allocation constraints: The variable is 0-1, representing whether unit i is in maintenance state at time t under scenario s. 0 indicates not in maintenance state, and 1 indicates in maintenance state. The maintenance personnel allocation constraint indicates that if unit i is in maintenance state at time t, i.e. Then maintenance personnel need to be assigned. ;

[0105] Maintenance personnel number constraints: The maintenance personnel number constraint indicates that at any given time there can be a maximum of [number missing] maintenance personnel. The maintenance team's work exceeds the time limit, triggering a high-cost penalty; the excess portion is used... Indicates temporary employment;

[0106] Unit operating status update: In the formula, To indicate a specific moment during short-term maintenance, It is a 0-1 variable, representing the time of unit i at time... Whether maintenance is started: 0 indicates maintenance is not started, 1 indicates maintenance is started. This is a 0-1 variable, representing whether group i is available on day d in scenario s, where 0 indicates unavailable and 1 indicates available. This represents the total number of days in the long-term maintenance phase. Representing a scene Lower unit exist Is the time available? (0 indicates available, 0 indicates unavailable); the numerator ensures the fan is available again after maintenance; the denominator avoids division by zero. This is to avoid extremely small positive numbers with a denominator of 0;

[0107] Constraints are unavailable during maintenance: During maintenance, constraints cannot be used to indicate that the wind turbine cannot generate electricity while it is under maintenance.

[0108] Vessel scheduling and leasing constraints: In the formula, For a sufficiently large positive number, the vessel scheduling and leasing constraint means that if there is a planned maintenance task on a certain day, a vessel must be leased. (The law ensures the logic holds true).

[0109] Overtime constraints for maintenance personnel: In the formula, the overtime constraint for maintenance personnel means that the total working hours ≤ normal working hours (B×W) + overtime working hours. +Over-limit penalty On the first day of LTH, priority should be given to handling unfinished tasks, including those with remaining time. The daily overtime work hour limit is Hour;

[0110] Power output constraints: In the formula, For the scene Lower unit exist Electricity generation at any given time (MW); Rated capacity of the wind turbine (MW); The normalized power coefficient for wind speed; The long-term wind speed normalized power coefficient of unit i on day d under scenario s is between 0 and 1; during preventive maintenance, power generation is deducted proportionally. This is for preventative maintenance.

[0111] Grid absorption constraints: The grid absorption constraint means that the total output power is less than or equal to the grid's absorption capacity. Let be the grid absorption ratio at time t under scenario s, which is between 0 and 1.

[0112] S4: Employing a rolling time-domain optimization algorithm, a two-stage stochastic mixed integer programming model is solved daily, and a maintenance scheduling plan is output. For details on the rolling time-domain optimization algorithm of this invention, please refer to [link to relevant documentation]. Figure 3 The specific steps are as follows:

[0113] S41. State initialization:

[0114] S411. Will Set to 1, A 0-1 variable, representing the unit The indicator shows whether the maintenance task has been completed; 1 indicates that it has not been completed, and 0 indicates that it has been completed.

[0115] S412. Will Set to 1, It is a 0-1 variable, indicating whether the unit has started maintenance, with 1 indicating that it has not started and 0 indicating that it has started; A 0-1 variable, representing the unit Whether it is available, 1 indicates available, 0 indicates unavailable;

[0116] S413. Set the current scroll wheel number Setting it to 0 indicates that the current scroll wheel number is the first scroll wheel;

[0117] S42. Enter the time-domain rolling loop;

[0118] S421. Set the short-term time window to... This indicates that the short-term time span of each round of rolling is 24 hours;

[0119] S422. Set the long-term time window to This indicates the long-term time span of each round of rolling. sky;

[0120] S43. Calculate state variables

[0121] Based on the already generated The scenarios include wind speed, wave height, electricity price, and remaining lifespan of generator components, combined with wind speed. High waves A safety threshold determines the duration of a maintenance task that can be performed, i.e., weather opportunity; based on wind speed. and electricity price Computerized maintenance of low power generation loss opportunities; determination of preventive or fail-safe maintenance based on the expected remaining lifespan of unit components;

[0122] S44. Solving a two-stage stochastic mixed integer programming model

[0123] The required data for the two-stage stochastic mixed integer programming model is input and the Gurobi solver is called to solve it, yielding the short-term scheduling variables. and long-term scheduling variables The values ​​are then combined to obtain the scheduling plan for this rolling solution;

[0124] S45. Status Update

[0125] Units that have completed maintenance tasks That Set to 0; units whose maintenance tasks have not been completed will have their maintenance tasks set to 0. Set the value to , Value represents the unit Remaining repair time when repairs are not completed in short-term scenarios; Set the value to 1. A 0-1 variable used to label the generator set. Whether it is "transferred" to the long-term maintenance phase, 1 indicates that it needs to be put into the long-term maintenance phase, and 0 indicates that it does not need to be put into the long-term maintenance phase;

[0126] S46. Rolling Iteration

[0127] Will Set as This indicates that the rolling iteration has entered the next round; and the life status of the wind turbine is updated based on the completion status of this maintenance task.

[0128] S47. Termination and Output

[0129] If all Then the algorithm will terminate rolling and all schedule results will be output as a maintenance plan.

[0130] The steps of the rolling time-domain optimization algorithm are as follows:

[0131] (1) Update the unit's maintenance status daily.

[0132] (2) Based on the collected data on wind speed, wave height, electricity price, and remaining lifespan of unit components, an uncertainty scenario generation model is constructed using a Gaussian process regression model and a Weibull distribution, and the model is generated. An uncertain scenario, then the computer group maintenance opportunity;

[0133] (3) Use the calculated data as input to the two-stage stochastic mixed integer programming model and call the Gurobi solver to solve the model;

[0134] (4) Integrate the solution results to construct a complete maintenance plan for the current iteration;

[0135] (5) Update the wind turbine status and model input data and use rolling iteration for the next round of optimization until all maintenance tasks are scheduled.

[0136] Real-world examples

[0137] The scheduling strategy of the present invention will be further illustrated below with a specific case.

[0138] Simulation examples and calculation parameters

[0139] This case study selects an actual wind farm as the scheduling target for this operation and maintenance (O&M) operation. Five wind turbines are scheduled for maintenance, with a scheduling period of 24 days. The scheduling is based on the maintenance resources available at this wind farm. Please refer to Tables 1, 2, and 3 for wind farm parameters, model parameters, and data sources.

[0140] Table 1. Basic parameters of the wind farm:

[0141] Table 2 Maintenance resource parameters:

[0142] Table 3 Environmental Data:

[0143] This case study sets up five scenarios based on existing scheduling strategies to verify the effectiveness of the offshore wind power operation and maintenance optimization method of this invention, which considers multi-source uncertainties and opportunistic collaborative maintenance. Scenario 1: Using a fault maintenance strategy, i.e., not considering uncertain scenarios and opportunistic collaboration, not considering long-term planning, and only performing maintenance after equipment failure and downtime; Scenario 2: Using a preventative maintenance strategy, i.e. not considering uncertain scenarios, not considering opportunistic collaboration, only performing maintenance based on the remaining lifespan of the unit, considering long-term overall planning rather than rolling optimization; Scenario 3: Using an opportunistic maintenance strategy, i.e. considering opportunistic collaborative maintenance, but not considering uncertain scenarios, considering long-term planning and using rolling optimization; Scenario 4: Using the strategy proposed in this invention, i.e. considering uncertain scenarios, considering opportunistic collaboration and rolling optimization; Scenario 5: Using a full-information overall opportunistic strategy, i.e. considering opportunistic collaboration, not considering uncertainties, but with no scenario prediction error.

[0144] First, according to such Figure 1 The illustrated implementation steps of the present invention, in each rolling cycle, are based on the latest acquired weather, electricity price, and monitoring system forecast information, and according to... Figure 2 The probabilistic prediction method shown generates 50 sets of scenario trajectories for wind speed, wave height, and electricity price, as well as the remaining lifespan of the generating unit, and calculates the opportunities for maintenance tasks, which are then used as input to a two-stage stochastic mixed-integer programming model. For the confidence intervals of the uncertain scenario trajectories for wind speed, wave height, and electricity price, please refer to [link to relevant documentation]. Figure 4 Please refer to the weather opportunity results for unit maintenance. Figure 5 In the figure, the red part represents the weather-inaccessible stage, the transparent part represents the weather-accessible stage, and the three bars represent the accessibility results of uncertain scenarios, the accessibility results of point prediction scenarios, and the actual accessibility results, respectively. It can be found that uncertain scenarios can avoid misjudgment of point prediction accessibility due to inaccurate predictions multiple times, which is of great significance for improving the robustness of the strategy and reducing operation and maintenance costs; the opportunity for low power generation loss is determined by the optimization model after calculating the operation and maintenance costs.

[0145] Then, according to such Figure 3The rolling time-domain optimization algorithm steps shown involve solving a two-stage stochastic mixed-integer linear programming model to obtain the daily maintenance plan. After completing the current rolling plan, it is rolled over to the next day for maintenance task planning until all maintenance tasks are completed. The daily maintenance plans obtained from each rolling cycle are integrated to form a complete maintenance plan. The comparison strategy has been re-encoded and recalculated through simulation in this case study. The maintenance plan results are available in the reference section. Figure 6 .analyze Figure 6 It can be seen that, apart from the strategy of this invention and the overall opportunity maintenance strategy of full information, the other three strategies all have maintenance interruptions. Due to the scheduling of maintenance resources, this will greatly increase the operation and maintenance cost. At the same time, it also reflects the improvement of the robustness and economy of the operation and maintenance strategy by considering the uncertainty scenario of this invention.

[0146] Finally, after conducting 100 experiments using different wind speeds, wave heights, and electricity price data, please refer to the box plot of the total cost of maintenance tasks in the five scenarios of this case study and the power generation profit curve of a certain experiment. Figure 7 .analyze Figure 7 (a) It can be seen that, on the one hand, the expected profit of the uncertain scenario is similar to that of the point prediction scenario, which reflects the rationality of the confidence interval for the uncertain scenario generated by this invention; on the other hand, the confidence interval for the profit of the uncertain scenario shows the characteristic of widening when the actual profit fluctuates greatly. This means that the uncertain scenario has an early warning function when weather and electricity prices fluctuate frequently, which will lead to an increase in expected costs in the optimization model, thus forming a tendency to avoid maintenance during this period. Analysis Figure 7 (b) Based on the statistical results of 100 experiments, the strategy proposed in this invention has the lowest median maintenance cost, the narrowest interquartile range, and good control over extreme values ​​among all strategies, demonstrating lower average cost, higher scheduling stability, and stronger robustness. A comparative analysis of the five strategies shows that:

[0147] (1) Compared with the fault maintenance strategy, the maintenance cost of the present invention is significantly lower. The fault maintenance strategy relies solely on the unit fault information for scheduled maintenance, lacks preventive maintenance thinking and the ability to coordinate opportunities, and its passivity easily leads to high downtime losses and resource waste.

[0148] (2) Compared with preventive maintenance strategies, this invention introduces uncertainty modeling and opportunity coordination mechanisms in the optimized structure, resulting in smaller cost dispersion and better economic efficiency. Because preventive maintenance strategies lack opportunity coordination and scenario adaptability, their median cost is higher and the fluctuation range is larger.

[0149] (3) Compared with the opportunity maintenance strategy, the present invention reduces the median cost by about 24%. Although the opportunity maintenance strategy in this case considers opportunity coordination, it only schedules based on the prediction of a single scene point and does not consider the uncertainty caused by the prediction error. It is easily affected by environmental disturbances, and its cost shows a larger fluctuation range and a higher worst-case cost than the strategy of the present invention.

[0150] (4) Although the overall opportunity scheduling strategy based on full information has a slightly higher cost than the strategy proposed in this invention, this method assumes that future information is completely and accurately known, which is not feasible in reality and can only serve as a theoretical optimal benchmark. In contrast, this invention does not rely on ideal information, has both feasibility and robustness, and is suitable for practical engineering applications.

[0151] In summary, the maintenance method proposed in this invention, which considers multi-source uncertainty and opportunity coordination, is significantly superior to existing scheduling strategies in terms of cost control, risk avoidance, and robustness. The above embodiments fully verify the significant advantages of this invention in unifying three types of maintenance opportunities, coping with multi-source uncertainty, and achieving coordinated optimization of economy and reliability, and have good promotional value.

Claims

1. A method for offshore wind power operation and maintenance that considers both uncertainty and opportunity in collaborative maintenance, characterized in that, The method includes the following steps: S1: Collects point prediction data on wind speed, wave height, electricity price, wind speed, wave height, electricity price, and predicted remaining lifespan of wind turbine components. ; S2: Generated through a spatiotemporal probability model A group of uncertain scenarios, among which Indicates the number of scenes: By combining wind speed and wave height uncertainties with weather maintenance thresholds, the weather opportunities for operation and maintenance are assessed; by combining wind speed-power generation relationship with electricity price uncertainty, the low power generation loss opportunities for operation and maintenance are assessed; based on the remaining lifespan of unit components, the maintenance method of preventive maintenance or fault maintenance is evaluated by the subsequent optimization model; the above results will be used as inputs for the subsequent optimization model. S3: Construct a two-stage stochastic mixed integer programming model as the optimization model, and set the decision variables and constraints of the model; S4: Employs a rolling time-domain optimization algorithm to solve a two-stage stochastic mixed integer programming model daily and output a maintenance scheduling plan.

2. The offshore wind power operation and maintenance method considering both uncertainty and opportunity in collaborative maintenance as described in claim 1 is characterized in that, The specific methods for step S2 include: S21. Scenario generation based on uncertain wind speed, wave height, and electricity price is based on: In the formula, For predictor variables, wind speed / wave height / electricity price at time t = 0. The value; For a priori wind speed / wave height / electric value; These are the residual values ​​calculated based on a zero-mean Gaussian process regression model; To satisfy the requirement of independent and identically distributed white noise, , To represent the noise variance, the joint distribution of uncertain scenarios for wind speed, wave height, and electricity price is obtained by training a Gaussian process regression model, and several uncertain scenarios are obtained through several random samplings. S22. Wind turbine remaining life prediction is based on: Assuming each unit The remaining lifetime follows a Weibull distribution: In the formula, To indicate the unit Random variables that predict remaining lifespan; To monitor the system and predict the lifespan of unit components, it serves as a scale parameter for the Weibull distribution; Let be the shape parameter of the Weibull distribution; by using Random sampling generates the remaining life values ​​of unit components under several uncertain scenarios, thereby determining whether to adopt preventive maintenance or fault maintenance for the unit under each scenario.

3. The offshore wind power operation and maintenance method considering both uncertainty and opportunity in collaborative maintenance as described in claim 2 is characterized in that... The steps for prediction based on the zero-mean Gaussian process regression model are as follows: The following steps are used to predict values ​​for future times. The model is constructed to model the distribution of the residuals, which implicitly includes the residual process. Spatiotemporal correlation: S211. Construct the covariance matrix , Size is , At the current time point, elements in for: In the formula, It is an indicator function, when The value is 1 if it is true, and 0 otherwise. It is the covariance function. It is the noise variance; S212. Selection of covariance function: Because wind speed, wave height, and electricity price have a strong time correlation, the covariance function... Selected as the squared exponent kernel: In the formula, For signal variance, For length scale, It is an exponential function; S213. Model training, i.e., hyperparameter estimation: Hyperparameters are estimated by maximizing marginal likelihood. : In the formula, These are historical observation values; The corresponding predicted value for the original point. Let be the logarithm of the marginal likelihood function. Let be the logarithm of the determinant of the covariance matrix. It is the natural logarithm function. Pi, with the superscript T representing matrix transpose. S214. Predicting the Future Step, i.e., moment value Its conditional distribution is a Gaussian distribution with a mean of: The variance is: In the formula, For the original point prediction model in The predicted value at any given time; for A vector whose elements are historical moments. With the predicted time Covariance: ; The covariance matrix is ​​estimated based on historical data; The signal variance of the kernel function, i.e. ; For expectation calculation; For the predicted time The variance; Variance operator; S215. Scene Generation: Generate a random trajectory for the next H steps. , i.e., scenario: In the formula, To predict the mean vector; To predict the covariance matrix, the size is , elements in for: In the formula, for Random trajectory at time, for Random trajectory at time, It is the covariance function; Then, by analyzing random trajectories Several samplings were conducted to obtain several sets of random scenarios with uncertainties in wind speed, wave height, and electricity price. By combining the uncertain scenarios of wind speed and wave height with weather maintenance thresholds, the weather opportunities for operation and maintenance were assessed. By combining the wind speed-power generation relationship with the uncertain scenarios of electricity price, the low power generation loss opportunities for operation and maintenance were assessed.

4. The offshore wind power operation and maintenance method considering both uncertainty and opportunity in collaborative maintenance as described in claim 1 is characterized in that... The objective function of the two-stage stochastic mixed-integer programming model described in step S3 is: In the formula, For short-term gains, i.e., intraday profits. For the first Long-term profits for the day To maintain the cycle number of days, The number of uncertain scenarios; A collection of uncertain scenarios; A collection of units awaiting maintenance; For the scene Next Individual unit maintenance interruption indicator variable, when hour ; For the scene Next Remaining time for each unit to complete its task; Penalty factor related to maintenance interruption; Penalty costs for temporarily hiring maintenance personnel and for staff working overtime; This is an auxiliary variable that exceeds the budget for the number of maintenance personnel and the budget for overtime hours; The variable is 0-1, indicating whether short-term routine maintenance will be initiated for unit i at time t, where 1 indicates maintenance will be initiated and 0 indicates maintenance will not be initiated. This is a 0-1 variable, indicating whether long-term maintenance is performed on unit i on day d in scenario s. This is a 0-1 variable representing whether a vessel is leased for short-term maintenance; 0 indicates no leasing, and 1 indicates leasing. The variable is 0-1, indicating whether to rent a boat on day d in scenario s, where 0 means not to rent and 1 means to rent. Short-term gains, i.e., intraday profits for: In the formula, Maintenance cycle duration; For the unit Running status This indicates that it is normal. Indicates a fault; For fault maintenance costs, To cover preventative maintenance costs, To determine whether maintenance began the previous day but was not completed, It means yes; For the unit Is it in STH? Maintenance is initiated at any time; 1 indicates maintenance is initiated, and 0 indicates maintenance is not performed. Regarding whether to charter a boat, Indicates leasing. Indicates that it is not rented; Daily rental fee for the vessel; To maintain the hourly wage of staff; Overtime pay for maintenance personnel (hourly rate); For the scene Are the maintenance personnel present? Constantly monitor the unit Performing maintenance (1 indicates maintenance is in progress, 0 indicates maintenance is not in progress); For the scene Overtime hours; For the scene Next moment Electricity price; For the scene Next moment unit Output power; No. Long-term profits for: In the formula, Let d be the long-term electricity price forecast for day d under scenario s. Let i be the long-term output power of unit i on day d under scenario s. The variable is 0-1, representing the long-term operating status of unit i on day d under scenario s, where 1 indicates normal operation and 0 indicates shutdown for maintenance. This is a 0-1 variable, representing whether long-term maintenance is performed on unit i on day d under scenario s. 0 indicates no long-term maintenance, and 1 indicates long-term maintenance. Let d be the time required for long-term maintenance of unit i on day d under scenario s, in hours. Let d be the number of long-term overtime hours on day d under scenario s; For fault maintenance costs, To cover preventative maintenance costs, To determine whether maintenance began the previous day but was not completed, Yes, Indicates no.

5. The offshore wind power operation and maintenance method considering both uncertainty and opportunity in collaborative maintenance as described in claim 1 is characterized in that... The constraints of the stochastic mixed-integer optimization model described in step S3 include: Task assignment uniqueness constraint: In the formula, represents each unit. Maintenance tasks must be scheduled in STH or LTH and only once; Daily work time window constraints: The daily work time window constraint means that maintenance tasks can only be performed during the day. to It can be done between times, and cannot be earlier than time. No later than the time ; Maintenance task continuity constraints: In the formula, To determine the time required to complete maintenance task i starting from time t in scenario s, the impact of weather accessibility is taken into account; the maintenance task continuity constraint means that after the task starts, the crew must remain in maintenance mode until it is completed or STH ends. Maintenance interruptions and intertemporal constraints: Maintenance interruption and time-series constraints indicate the remaining time if a task is not completed in STH. It must continue at LTH and trigger the interruption penalty at this time. ; Maintenance personnel allocation constraints: The variable is 0-1, representing whether unit i is in maintenance state at time t under scenario s. 0 indicates not in maintenance state, and 1 indicates in maintenance state. The maintenance personnel allocation constraint indicates that if unit i is in maintenance state at time t, i.e. Then maintenance personnel need to be assigned. ; Maintenance personnel number constraints: The maintenance personnel number constraint indicates that at any given time there can be a maximum of [number missing] maintenance personnel. The maintenance team's work exceeds the time limit, triggering a high-cost penalty; the excess portion is used... Indicates temporary employment; Unit operating status update: In the formula, To indicate a specific moment during short-term maintenance, It is a 0-1 variable, representing the time of unit i at time... Whether maintenance is started: 0 indicates maintenance is not started, 1 indicates maintenance is started. This is a 0-1 variable, representing whether group i is available on day d in scenario s, where 0 indicates unavailable and 1 indicates available. This represents the total number of days in the long-term maintenance phase. Representing a scene Lower unit exist Is the time available? (0 indicates available, 0 indicates unavailable); the numerator ensures the fan is available again after maintenance; the denominator avoids division by zero. This is to avoid extremely small positive numbers with a denominator of 0; Constraints are unavailable during maintenance: During maintenance, constraints cannot be used to indicate that the wind turbine cannot generate electricity while it is under maintenance. Vessel scheduling and leasing constraints: In the formula, For a sufficiently large positive number, the vessel scheduling and leasing constraint means that if there is a planned maintenance task on a certain day, a vessel must be leased. (The law ensures the logic holds true). Overtime constraints for maintenance personnel: In the formula, the overtime constraint for maintenance personnel means that the total working hours ≤ normal working hours (B×W) + overtime working hours. +Over-limit penalty On the first day of LTH, priority should be given to handling unfinished tasks, including those with remaining time. The daily overtime work hour limit is Hour; Power output constraints: In the formula, For the scene Lower unit exist Electricity generation at any given time (MW); Rated capacity of the wind turbine (MW); The normalized power coefficient for wind speed; The long-term wind speed normalized power coefficient of unit i on day d under scenario s is between 0 and 1; during preventive maintenance, power generation is deducted proportionally. This is for preventative maintenance; Grid absorption constraints: The grid absorption constraint means that the total output power is less than or equal to the grid's absorption capacity. Let be the grid absorption ratio at time t under scenario s, which is between 0 and 1.

6. The offshore wind power operation and maintenance method considering both uncertainty and opportunity in collaborative maintenance as described in claim 1 is characterized in that, The specific steps of the rolling time-domain algorithm described in step S4 are as follows: S41. State initialization: S411. Will Set to 1, A 0-1 variable, representing the unit The indicator shows whether the maintenance task has been completed; 1 indicates that it has not been completed, and 0 indicates that it has been completed. S412. Will Set to 1, It is a 0-1 variable, indicating whether the unit has started maintenance, with 1 indicating that it has not started and 0 indicating that it has started; A 0-1 variable, representing the unit Whether it is available, 1 indicates available, 0 indicates unavailable; S413. Set the current scroll wheel number Setting it to 0 indicates that the current scroll wheel number is the first scroll wheel; S42. Enter the time-domain rolling loop; S421. Set the short-term time window to... This indicates that the short-term time span of each round of rolling is 24 hours; S422. Set the long-term time window to This indicates the long-term time span of each round of rolling. sky; S43. Calculate state variables Based on the already generated The scenarios include wind speed, wave height, electricity price, and remaining lifespan of generator components, combined with wind speed. High waves A safety threshold determines the duration of a maintenance task that can be performed, i.e., weather opportunity; based on wind speed. and electricity price Computerized maintenance of low power generation loss opportunities; determination of preventive or fail-safe maintenance based on the expected remaining lifespan of unit components; S44. Solving a two-stage stochastic mixed integer programming model The required data for the two-stage stochastic mixed integer programming model is input and the Gurobi solver is called to solve it, yielding the short-term scheduling variables. and long-term scheduling variables The values ​​are then combined to obtain the scheduling plan for this rolling solution; S45. Status Update Units that have completed maintenance tasks That Set to 0; units whose maintenance tasks have not been completed will have their maintenance tasks set to 0. Set the value to , Value represents the unit Remaining repair time when repairs are not completed in short-term scenarios; Set the value to 1. A 0-1 variable used to label the generator set. Whether it is "transferred" to the long-term maintenance phase, 1 indicates that it needs to be put into the long-term maintenance phase, and 0 indicates that it does not need to be put into the long-term maintenance phase; S46. Rolling Iteration Will Set as This indicates that the rolling iteration has entered the next round; and the life status of the wind turbine is updated based on the completion status of this maintenance task. S47. Termination and Output If all Then the algorithm will terminate and all schedule results will be output as a maintenance plan; The steps of the rolling time-domain optimization algorithm are as follows: (1) Update the unit's maintenance status daily. (2) Based on the collected data on wind speed, wave height, electricity price, and remaining lifespan of unit components, an uncertainty scenario generation model is constructed using a Gaussian process regression model and a Weibull distribution, and the model is generated. An uncertain scenario, then the computer group maintenance opportunity; (3) Use the calculated data as input to the two-stage stochastic mixed integer programming model and call the Gurobi solver to solve the model; (4) Integrate the solution results to construct a complete maintenance plan for the current iteration; (5) Update the wind turbine status and model input data and use rolling iteration for the next round of optimization until all maintenance tasks are scheduled.

Images