Coordination control method and system for offshore wind plant and storage medium

Abstract

The invention relates to the technical field of wind power plant control, in particular to a coordination control method and system for an offshore wind power plant and a storage medium. The method comprises the following steps: firstly, obtaining theoretical maximum active power according to a wind speed predicted value of each wind turbine generator, and obtaining an active power demand upper limit and a reactive power range of a wind power plant in combination with power grid dispatching data and actual output power of the wind power plant; then, obtaining a power reference value of each wind turbine generator by using a multi-objective optimization power prediction model and an improved particle swarm algorithm; in combination with the power reference value and the fan state information, a reinforcement learning algorithm is utilized to optimize group groups and power distribution in the groups, and an actual power reference value is obtained; thirdly, obtaining an actual power reference correction value by using a power compensation correction formula; and finally, controlling the active power and the reactive power of each wind turbine generator by adopting maximum power point tracking control according to the actual power reference correction value. The effectiveness of wind power plant coordination control can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of wind farm control, and in particular to a coordinated control method, system and storage medium for an offshore wind farm. Background Art

[0002] The coordinated control method of offshore wind farms is to send the power reference value calculated by the wind farm controller to each wind turbine and compare it with the actual value of the wind turbine, so as to use the wind turbine controller to adjust the power difference to achieve the power reference value; at the same time, the power reference value can also be converted into other parameter changes through gain, such as voltage and current, to adjust the wind turbine corresponding to the wind farm parameters.

[0003] At present, the existing technology still has shortcomings in the coordinated control of offshore wind farms. On the one hand, the reference values ​​obtained by the wind farm control equipment cannot adapt to the changes in the operating status of the power grid and the wind farm, which may cause grid voltage fluctuations or wind farm power factor failure to meet the standards. On the other hand, the existing technology does not adaptively adjust the group division and power allocation of wind turbines according to the real-time operating status of the wind farm and the needs of the power grid, which will reduce the effectiveness of coordinated control of offshore wind farms.

[0004] To this end, a coordinated control method, system and storage medium for an offshore wind farm are proposed. Summary of the invention

[0005] The purpose of the present invention is to provide a coordinated control method, system and storage medium for an offshore wind farm. First, according to the wind speed prediction value of each wind turbine, the theoretical maximum active power is obtained, and combined with the grid dispatching data and the actual output power of the wind farm, the upper limit of the active power demand and the reactive power range of the wind farm are obtained; then, the power reference value of each wind turbine is obtained by using a multi-objective optimization power prediction model and an improved particle swarm algorithm; combined with the power reference value and the wind turbine status information, the group grouping and the power distribution within the group are optimized by using a reinforcement learning algorithm to obtain the actual power reference value; then, the actual power reference correction value is obtained by using a power compensation correction formula; finally, according to the actual power reference correction value, the maximum power point tracking control is used to control the active power and reactive power of each wind turbine.

[0006] To achieve the above object, the present invention provides the following technical solutions: A coordinated control method for an offshore wind farm, comprising: The short-term wind speed prediction model based on Kalman filter is used to obtain the wind speed prediction value of each wind turbine; According to the wind speed prediction value, the theoretical maximum active power of each wind turbine is obtained; and the upper limit of the active power demand and the reactive power range of the wind farm are obtained by combining the theoretical maximum active power, the grid dispatching data and the actual output power of the wind farm; Constructing a multi-objective optimization power prediction model, combining the active power demand upper limit and the reactive power range, and using an improved particle swarm algorithm to solve the multi-objective optimization power prediction model to obtain a power reference value for each wind turbine; Combining the power reference value and wind turbine status information, using a reinforcement learning algorithm to optimize grouping and power allocation within the group, to obtain an actual power reference value for each wind turbine set; Inputting the actual power reference value of each wind turbine generator set into a power compensation correction formula to obtain an actual power reference correction value of each wind turbine generator set; According to the actual power reference correction value, the active power and reactive power of each wind turbine generator set are controlled by adopting the maximum power point tracking control.

[0007] Furthermore, the process of obtaining the upper limit of the active power demand and the reactive power range of the wind farm includes: Obtaining the maximum allowable power and power factor range from the power grid dispatching data; Combining the wind speed prediction value with the wind turbine power curve, obtaining the theoretical maximum active power of each wind turbine; The theoretical maximum active power of each wind turbine is summed to obtain the theoretical maximum active power of the wind farm; Comparing the maximum allowable power with the theoretical maximum active power of the wind farm to obtain the upper limit of the active power demand; The reactive power range is obtained by combining the power factor range with the actual output power of the wind farm.

[0008] Furthermore, the process of obtaining the power reference value of each wind turbine generator set includes: The total power generation income of the wind farm, the life loss of the wind turbines and the voltage deviation of the wind farm grid connection point are taken as the optimization objectives of the multi-objective optimization power prediction model and the total objective function is set; Setting the active power demand upper limit and the reactive power range as wind farm total power constraint conditions of the multi-objective optimization power prediction model; The multi-objective optimization power prediction model is solved by using an improved particle swarm algorithm to obtain the power reference value of each wind turbine generator set.

[0009] Furthermore, the process of obtaining the actual power reference value of each wind turbine generator set includes: Define the state space, action space, and reward function of a Markov decision process; The state space includes: the wind turbine state information, group information and the power reference value; the action space includes: the grouping matrix, the active power allocation coefficient and the reactive power allocation coefficient; the reward function includes: power deviation reward, average mechanical load reward and group quantity change reward; The Actor network and the Critic network are trained using deep deterministic policy gradients and historical data, and the current wind turbine state information and the current power reference value are input into the trained network to obtain the actual power reference value.

[0010] Furthermore, the power compensation correction formula is expressed as: ; in, Indicates Wind turbines in The actual power reference correction value at the moment; Indicates the minimum value operation; Indicates Wind turbines in The actual power reference value at the moment; Indicates the wind speed compensation coefficient; Indicates Wind turbines in The wind speed forecast value at time; wherein, is the communication delay time; Indicates Wind turbines in The current wind speed value at the moment; represents the delay compensation coefficient; represents the estimated value of power variation due to communication delay; Indicates Wind turbines in The available power at the moment.

[0011] A coordinated control system for an offshore wind farm, comprising: a system control module, a data acquisition module, a multi-objective optimization power prediction module, a group power allocation module, a power compensation correction module and a power control module; The system control module is used to control the start, pause and stop of the system; The data acquisition module is used to obtain the wind speed forecast value and wind turbine status information of each wind turbine group and the upper limit of active power demand and reactive power range of the wind farm; The multi-objective optimization power prediction module is used to combine the active power demand upper limit and the reactive power range, and use an improved particle swarm algorithm to solve the multi-objective optimization power prediction model to obtain a power reference value for each wind turbine; The group power allocation module is used to combine the power reference value and the wind turbine status information, and optimize the grouping and power allocation within the group by using a reinforcement learning algorithm to obtain an actual power reference value for each wind turbine group; The power compensation correction module is used to input the actual power reference value of each wind turbine group into the power compensation correction formula to obtain the actual power reference correction value of each wind turbine group; The power control module is used to control the active power and reactive power of each wind turbine generator set by adopting maximum power point tracking control according to the actual power reference correction value.

[0012] A storage medium stores computer program instructions, and when the computer program is executed by a processor, the computer program implements the steps in a coordinated control method for an offshore wind farm.

[0013] Compared with the prior art, the present invention has the following beneficial effects: 1. The present invention proposes a multi-objective optimization power prediction method for obtaining the global optimal power reference value of a wind farm; the method realizes power optimization by constructing the objective function and constraint conditions of a multi-objective optimization power prediction model; the model takes into account the power generation income, life loss and grid connection point voltage of the wind farm, and is solved in combination with an improved particle swarm algorithm to obtain the power reference value of each wind turbine under the global optimal state; the process optimizes power from the overall perspective of the wind farm and the power grid, and provides global reference data for subsequent local power optimization, thereby effectively improving the effectiveness of coordinated control of offshore wind farms.

[0014] 2. The present invention proposes a group power allocation method for obtaining the local optimal actual power reference value of a wind turbine group; the method obtains the actual power reference value by defining the state space, action space and reward function of the Markov decision process and combining the deep deterministic policy gradient algorithm. This process optimizes both the wind turbine grouping and the power allocation according to the state information of the unit, and can improve the effectiveness of the coordinated control of offshore wind farms by considering the local differences and actual constraints of the units.

[0015] 3. The present invention proposes a power compensation correction method for compensating and correcting power; the method inputs the actual power reference value into the power compensation correction formula to obtain the actual power reference correction value; the power compensation correction formula takes into account the actual impact of wind speed deviation, communication delay and available capacity on power, so that the actual power reference correction value can be more in line with the actual situation, thereby improving the effectiveness of coordinated control of offshore wind farms. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A schematic flow chart of a coordinated control method for an offshore wind farm according to the present invention; Figure 2 It is a schematic structural diagram of a coordinated control system of an offshore wind farm according to the present invention. DETAILED DESCRIPTION

[0017] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0018] See also Figure 1 and Figure 2 The present invention provides a coordinated control method, system and storage medium for an offshore wind farm, and the technical solution is as follows:

[0019] Embodiment 1 A power plant uses a coordinated control method for an offshore wind farm proposed by the present invention to effectively ensure the power stability of each wind turbine in the offshore wind farm. The flow chart of the method is as follows: Figure 1 As shown, the details are as follows: The short-term wind speed prediction model based on Kalman filter is used to obtain the wind speed prediction value of each wind turbine; Furthermore, the process of obtaining the wind speed prediction value of each wind turbine generator set includes: Call the historical wind speed data of each wind turbine from the background monitoring system; Perform data cleaning and preprocessing on historical wind speed data to obtain training data; Construct a short-term wind speed prediction model based on Kalman filtering, and define the state variables, state transfer equations, observation equations and noise covariance matrix of the Kalman filter; the state variables include wind speed and wind speed change rate; The Kalman filter is iteratively trained using the training data, and the trained Kalman filter is used to perform short-term wind speed forecasting to obtain the wind speed forecast value of each wind turbine.

[0020] According to the wind speed forecast value, the theoretical maximum active power of each wind turbine is obtained; the upper limit of the active power demand and the reactive power range of the wind farm are obtained by combining the theoretical maximum active power, the grid dispatching data and the actual output power of the wind farm; Furthermore, the process of obtaining the upper limit of active power demand and reactive power range of the wind farm includes: Obtain the maximum allowable power and power factor range from the power grid dispatch data; Combining the wind speed prediction value and the wind turbine power curve, the theoretical maximum active power of each wind turbine is obtained; The theoretical maximum active power of each wind turbine is summed to obtain the theoretical maximum active power of the wind farm; Compare the maximum allowable power with the theoretical maximum active power of the wind farm to obtain the upper limit of active power demand; Combining the power factor range with the actual output power of the wind farm, the reactive power range is obtained; Furthermore, the grid dispatch data and the actual output power of the wind farm are obtained by calling from the background monitoring system; Furthermore, the wind turbine power curve is obtained by using a wind speed-power lookup table.

[0021] The following are the upper limits of active power demand and reactive power ranges of wind farms obtained at three different times, which are respectively recorded as sample 1, sample 2 and sample 3, as shown in Table 1.

[0022] Table 1 Active power demand upper limit and reactive power range

[0023] By obtaining the upper limit of active power demand and reactive power range of the wind farm based on wind speed forecast, grid dispatch data and actual output power, accurate constraint data can be provided for subsequent multi-objective optimization power forecasting, thereby accurately obtaining the theoretical power reference value of each wind turbine and further improving the effectiveness of coordinated control of offshore wind farms.

[0024] Construct a multi-objective optimization power prediction model, combine the upper limit of active power demand and reactive power range, and use the improved particle swarm algorithm to solve the multi-objective optimization power prediction model to obtain the power reference value of each wind turbine; Furthermore, the process of obtaining the power reference value of each wind turbine generator set includes: The total power generation income of the wind farm, the life loss of the wind turbines and the voltage deviation of the wind farm grid connection point are taken as the optimization objectives of the multi-objective optimization power prediction model and the total objective function is set; The upper limit of active power demand and reactive power range are set as the total power constraints of the wind farm in the multi-objective optimization power prediction model; The improved particle swarm algorithm is used to solve the multi-objective optimization power prediction model to obtain the power reference value of each wind turbine; Furthermore, the expression of the overall objective function is: ; in, represents the overall objective function; represents the total power generation revenue weight; represents the total power generation revenue objective function of the wind farm; represents the life loss weight; represents the objective function of wind turbine life loss; Indicates the grid connection point voltage deviation weight; It represents the voltage deviation objective function of the wind farm grid connection point.

[0025] Furthermore, the total power generation revenue weight, life loss weight and grid connection point voltage deviation weight are set to 0.35, 0.3 and 0.35 respectively; of course, these weights can be flexibly set according to actual needs.

[0026] Furthermore, the objective function of the total power generation revenue of the wind farm can be expressed as: ; in, represents the total power generation revenue objective function of the wind farm; represents the maximization function; Indicates the number of wind turbines; express The electricity price at the time; Indicates Wind turbines in The actual active power at the moment; Indicates a time interval; Furthermore, the objective function of wind turbine life loss can be expressed as: ; in, represents the objective function of wind turbine life loss; represents the minimization function; Indicates the number of wind turbines; Indicates Wind turbines in The damage value at the moment can be obtained through the wind turbine load analysis model; Indicates Wind turbines in The actual active power at the moment; Indicates a time interval; Furthermore, the voltage deviation objective function of the wind farm grid connection point can be expressed as: ; in, represents the voltage deviation objective function of the wind farm grid connection point; represents the minimization function; Indicates the number of wind turbines; express The voltage at the wind farm grid connection point at the moment can be obtained through the power system flow calculation; express Reference value of grid connection point voltage at all times; Indicates a time interval; Furthermore, the improved particle swarm algorithm improves the convergence speed and accuracy of the algorithm by adaptively adjusting the parameters of the particle swarm algorithm, such as inertia weight, acceleration coefficient, etc.; the adaptive adjustment of the inertia weight can adopt a linear decreasing strategy, a nonlinear decreasing strategy and an adjustment strategy based on population diversity indicators; the adaptive adjustment of the acceleration coefficient adopts a dynamic adjustment strategy to dynamically balance the influence of the acceleration coefficient at different stages.

[0027] By constructing a multi-objective optimization power prediction model from the overall perspective of wind farms and power grids and using an improved particle swarm algorithm to obtain the global optimal solution, global reference data can be provided for subsequent local power optimization, thereby effectively improving the effectiveness of coordinated control of offshore wind farms.

[0028] Combining the power reference value and wind turbine status information, the reinforcement learning algorithm is used to optimize the grouping and power allocation within the group to obtain the actual power reference value of each wind turbine; Furthermore, the process of obtaining the actual power reference value of each wind turbine generator set includes: Define the state space, action space, and reward function of a Markov decision process; The state space includes: wind turbine state information, group information and power reference value; the action space includes: clustering matrix, active power allocation coefficient and reactive power allocation coefficient; the reward function includes: power deviation reward, average mechanical load reward and group quantity change reward; Furthermore, the wind turbine status information includes: the geographical location of the i-th wind turbine group and the actual wind speed, available useful power, available reactive power and mechanical load at time t; Furthermore, the group information includes: the current number of groups and a group member indication matrix; the group member indication matrix is ​​used to indicate the affiliation relationship of the wind turbines with respect to the group, for example: Indicates The wind turbine belongs to groups; Furthermore, the power reference value is the global optimal power allocated to each wind turbine by using a multi-objective optimization power prediction model, including an active power reference value and a reactive power reference value; Furthermore, the clustering matrix is ​​used to represent the clustering of wind turbines, including the probability of wind turbines being grouped; the active power allocation coefficient and the reactive power allocation coefficient are used to represent the power allocation action, and the sum of the active power allocation coefficient and the reactive power allocation coefficient of each group is 1; Furthermore, the reward function can be expressed as: ; in, represents the reward function; represents the power bias reward weight; represents the power deviation reward function, which is used to calculate the deviation between the actual power and the reference power; represents the average mechanical load reward weight; represents the mechanical load reward function, which is used to calculate the deviation between the mechanical load of a single unit and the average mechanical load of the total unit; Indicates the reward weight for changes in the number of groups; Represents the reward function for group number change, which is used to reduce the change in the number of groups; Furthermore, the power deviation reward function can be expressed as: ; in, represents the power deviation reward function; Indicates the number of wind turbines; Indicates Wind turbines in The actual active power at the moment; Indicates Wind turbines in Reference active power at the moment; Indicates Wind turbines in The actual reactive power at the moment; Indicates Wind turbines in Reference reactive power at the moment; Furthermore, the mechanical load reward function can be expressed as: ; in, represents the mechanical load reward function; Indicates the number of wind turbines; Indicates Wind turbines in Mechanical load at any time; Indicates that all wind turbines are Average mechanical load at the time; Furthermore, the reward function for group quantity change can be expressed as: ; in, Represents the reward function for changes in the number of groups; Indicates the penalty coefficient used to control the frequency of changes in the number of groups, which can be flexibly set according to actual needs; Indicates group Number of changes within Furthermore, the power deviation reward weight, the average mechanical load reward weight and the group quantity change reward weight are set to 0.35, 0.35 and 0.3 respectively; the settings of these weights can be adjusted according to actual conditions.

[0029] The Actor network and the Critic network are trained using a deep deterministic policy gradient algorithm and historical data, and the current wind turbine status information and the current power reference value are input into the trained network to obtain the actual power reference value.

[0030] The actual power reference value is obtained by defining the state space, action space and reward function of the Markov decision process and combining it with the deep deterministic policy gradient algorithm. This process not only optimizes the wind turbine grouping but also optimizes the power allocation according to the state information of the unit. By considering the local differences and actual constraints of the units, the effectiveness of coordinated control of offshore wind farms can be improved.

[0031] Inputting the actual power reference value of each wind turbine generator set into the power compensation correction formula to obtain the actual power reference correction value of each wind turbine generator set; Furthermore, the power compensation correction formula is expressed as: ; in, Indicates Wind turbines in The actual power reference correction value at the moment; Indicates the minimum value operation; Indicates Wind turbines in The actual power reference value at the moment; Indicates the wind speed compensation coefficient; Indicates Wind turbines in The wind speed forecast value at the time; is the communication delay time; Indicates Wind turbines in The current wind speed value at the moment; represents the delay compensation coefficient; represents the estimated value of power variation due to communication delay; Indicates Wind turbines in The available power at the moment.

[0032] Furthermore, considering the different dynamic characteristics of wind turbines and control systems, the wind speed compensation coefficient and the delay compensation coefficient need to be set according to actual conditions and are not unique.

[0033] In order to illustrate the effectiveness of the power compensation correction formula proposed in the present invention, three actual power reference values ​​and available active powers from different wind turbines at different times are randomly selected for power compensation correction tests, which are respectively recorded as test one, test two and test three; the corresponding wind speed compensation coefficient and delay compensation coefficient are set according to different units; the actual power reference correction value of each group is obtained in combination with the calculation process of the power compensation correction formula; wherein the actual power reference value includes the actual active power reference value (left side) and the actual reactive power reference value (right side); the actual power reference correction value includes the actual active power reference correction value (left side) and the actual reactive power reference correction value (right side); the power compensation correction test results are shown in Table 2.

[0034] Table 2 Power compensation correction test results

[0035] By considering the actual impact of wind speed deviation, communication delay and available capacity on power, the actual power reference correction value can be made more in line with the actual situation, thereby improving the effectiveness of coordinated control of offshore wind farms.

[0036] According to the actual power reference correction value, the maximum power point tracking control is adopted to control the active power and reactive power of each wind turbine.

[0037] This embodiment proposes a coordinated control method for an offshore wind farm. The method first combines the upper limit of the active power demand and the reactive power range of the wind farm and uses a multi-objective optimization power prediction model and an improved particle swarm algorithm to obtain a power reference value for each wind turbine; then, based on the power reference value and the wind turbine status information, a reinforcement learning algorithm is used to optimize the grouping and power allocation within the group to obtain an actual power reference value; then, the actual power reference correction value is obtained using a power compensation correction formula; finally, based on the actual power reference correction value, the maximum power point tracking control is used to control the active power and reactive power of each wind turbine. The method adopts a global-local-compensation correction strategy to improve the effectiveness of coordinated control of offshore wind farms.

[0038] Embodiment 2 The present invention proposes a coordinated control system for an offshore wind farm. The structure of the system can be referred to Figure 2 , including: system control module, data acquisition module, multi-objective optimization power prediction module, group power allocation module, power compensation correction module and power control module; The system control module is used to control the start, pause and stop of the system; The data acquisition module is used to obtain the wind speed forecast value and wind turbine status information of each wind turbine group, as well as the upper limit of active power demand and reactive power range of the wind farm; The multi-objective optimization power prediction module is used to combine the upper limit of active power demand and the reactive power range, and use the improved particle swarm algorithm to solve the multi-objective optimization power prediction model to obtain the power reference value of each wind turbine; Furthermore, the process of obtaining the power reference value of each wind turbine generator set by the multi-objective optimization power prediction module includes: The total power generation income of the wind farm, the life loss of the wind turbines and the voltage deviation of the wind farm grid connection point are taken as the optimization objectives of the multi-objective optimization power prediction model and the total objective function is set; The upper limit of active power demand and reactive power range are set as the total power constraints of the wind farm in the multi-objective optimization power prediction model; The improved particle swarm algorithm is used to solve the multi-objective optimization power prediction model to obtain the power reference value of each wind turbine.

[0039] The group power allocation module is used to combine the power reference value and wind turbine status information, and use the reinforcement learning algorithm to optimize the group grouping and power allocation within the group to obtain the actual power reference value of each wind turbine group; Furthermore, the process of the group power allocation module obtaining the actual power reference value of each wind turbine generator set includes: Define the state space, action space, and reward function of a Markov decision process; The state space includes: wind turbine state information, group information and power reference value; the action space includes: clustering matrix, active power allocation coefficient and reactive power allocation coefficient; the reward function includes: power deviation reward, average mechanical load reward and group quantity change reward; The Actor network and the Critic network are trained using deep deterministic policy gradients and historical data, and the current wind turbine status information and the current power reference value are input into the trained network to obtain the actual power reference value.

[0040] The power compensation correction module is used to input the actual power reference value of each wind turbine group into the power compensation correction formula to obtain the actual power reference correction value of each wind turbine group; Furthermore, the power compensation correction formula is expressed as: ; in, Indicates Wind turbines in The actual power reference correction value at the moment; Indicates the minimum value operation; Indicates Wind turbines in The actual power reference value at the moment; Indicates the wind speed compensation coefficient; Indicates Wind turbines in The wind speed forecast value at the time; is the communication delay time; Indicates Wind turbines in The current wind speed value at the moment; represents the delay compensation coefficient; represents the estimated value of power variation due to communication delay; Indicates Wind turbines in The available power at the moment.

[0041] The power control module is used to control the active power and reactive power of each wind turbine generator set by adopting maximum power point tracking control according to the actual power reference correction value.

[0042] The system of the present invention adopts a global-local-compensation correction scheme. In order to further verify the effectiveness of the scheme proposed by the present invention, the same test data is randomly selected from the data acquisition module to test and verify different schemes, which are respectively recorded as test one, test two and test three; the test scheme specifically includes: the scheme proposed by the present invention, that is, the multi-objective optimization power prediction module, the group power allocation module and the power compensation correction module, which is recorded as scheme one; the group power allocation module in the system is removed, and the multi-objective global optimization and compensation correction strategy is retained, which is recorded as scheme two; only the multi-objective optimization power prediction module in the system is retained, and the group local optimization and compensation correction strategy are removed, which is recorded as scheme three; The test data is input into different schemes to obtain the final power reference values ​​of the wind turbines under different schemes, and the proportion of the final power reference values ​​within a reasonable range is obtained through manual verification. The scheme comparison test results can be referred to Table 3.

[0043] Table 3 Comparative test results of the schemes

[0044] As shown in Table 3, compared with other solutions, the solution proposed in the present invention, namely Solution 1, can allocate a more reasonable power reference value to each wind turbine set, thereby effectively improving the power control of the offshore wind farm.

[0045] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A coordinated control method for an offshore wind farm, characterized in that: include: The short-term wind speed prediction model based on Kalman filter is used to obtain the wind speed prediction value of each wind turbine; According to the wind speed prediction value, the theoretical maximum active power of each wind turbine is obtained; and the upper limit of the active power demand and the reactive power range of the wind farm are obtained by combining the theoretical maximum active power, the grid dispatching data and the actual output power of the wind farm; Constructing a multi-objective optimization power prediction model, combining the active power demand upper limit and the reactive power range, and using an improved particle swarm algorithm to solve the multi-objective optimization power prediction model to obtain a power reference value for each wind turbine; Combining the power reference value and wind turbine status information, using a reinforcement learning algorithm to optimize grouping and power allocation within the group, to obtain an actual power reference value for each wind turbine set; Inputting the actual power reference value of each wind turbine generator set into a power compensation correction formula to obtain an actual power reference correction value of each wind turbine generator set; According to the actual power reference correction value, the active power and reactive power of each wind turbine generator set are controlled by adopting the maximum power point tracking control.

2. A coordinated control method for an offshore wind farm according to claim 1, characterized in that: The process of obtaining the upper limit of the active power demand and the reactive power range of the wind farm includes: Obtaining the maximum allowable power and power factor range from the power grid dispatching data; Combining the wind speed prediction value with the wind turbine power curve, obtaining the theoretical maximum active power of each wind turbine; The theoretical maximum active power of each wind turbine is summed to obtain the theoretical maximum active power of the wind farm; Comparing the maximum allowable power with the theoretical maximum active power of the wind farm to obtain the upper limit of the active power demand; The reactive power range is obtained by combining the power factor range with the actual output power of the wind farm.

3. A coordinated control method for an offshore wind farm according to claim 1, characterized in that: The process of obtaining the power reference value of each wind turbine generator set includes: The total power generation income of the wind farm, the life loss of the wind turbines and the voltage deviation of the wind farm grid connection point are taken as the optimization objectives of the multi-objective optimization power prediction model and the total objective function is set; Setting the active power demand upper limit and the reactive power range as wind farm total power constraint conditions of the multi-objective optimization power prediction model; The multi-objective optimization power prediction model is solved by using an improved particle swarm algorithm to obtain the power reference value of each wind turbine generator set.

4. A coordinated control method for an offshore wind farm according to claim 1, characterized in that: The process of obtaining the actual power reference value of each wind turbine generator set includes: Define the state space, action space, and reward function of a Markov decision process; The state space includes: the wind turbine state information, group information and the power reference value; the action space includes: the grouping matrix, the active power allocation coefficient and the reactive power allocation coefficient; the reward function includes: power deviation reward, average mechanical load reward and group quantity change reward; The Actor network and the Critic network are trained using deep deterministic policy gradients and historical data, and the current wind turbine state information and the current power reference value are input into the trained network to obtain the actual power reference value.

5. The coordinated control method of an offshore wind farm according to claim 1, characterized in that: The power compensation correction formula is expressed as: ； in, Indicates Wind turbines in The actual power reference correction value at the moment; Indicates the minimum value operation; Indicates Wind turbines in The actual power reference value at the moment; Indicates the wind speed compensation coefficient; Indicates Wind turbines in The wind speed forecast value at time; wherein, is the communication delay time; Indicates Wind turbines in The current wind speed value at the moment; represents the delay compensation coefficient; represents the estimated value of power variation due to communication delay; Indicates Wind turbines in The available power at the moment.

6. A coordinated control system for an offshore wind farm, characterized in that: include: System control module, data acquisition module, multi-objective optimization power prediction module, group power allocation module, power compensation correction module and power control module; The system control module is used to control the start, pause and stop of the system; The data acquisition module is used to obtain the wind speed forecast value and wind turbine status information of each wind turbine group and the upper limit of active power demand and reactive power range of the wind farm; The multi-objective optimization power prediction module is used to combine the active power demand upper limit and the reactive power range, and use an improved particle swarm algorithm to solve the multi-objective optimization power prediction model to obtain a power reference value for each wind turbine; The group power allocation module is used to combine the power reference value and the wind turbine status information, and optimize the grouping and power allocation within the group by using a reinforcement learning algorithm to obtain an actual power reference value for each wind turbine group; The power compensation correction module is used to input the actual power reference value of each wind turbine group into the power compensation correction formula to obtain the actual power reference correction value of each wind turbine group; The power control module is used to control the active power and reactive power of each wind turbine generator set by adopting maximum power point tracking control according to the actual power reference correction value.

7. A storage medium having computer program instructions stored thereon, characterized in that: When the computer program is executed by a processor, the steps in the coordinated control method of an offshore wind farm as claimed in any one of claims 1 to 5 are implemented.

Images