Wind power plant power reliable scheduling method based on vibration acceleration prediction
Through the optimization of vibration acceleration prediction model and evolutionary algorithm based on deep neural network, the problem of insufficient safety state judgment in wind farm scheduling is solved, and the accuracy and reliability of wind farm power scheduling is improved.
Patent Information
- Application Number
- CN202510761420.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-09
- Publication Date
- 2025-07-22
AI Technical Summary
The existing wind farms fail to accurately judge the safety status of the wind turbine during power scheduling, resulting in inaccurate and unreliable dispatch.
By establishing a vibration acceleration prediction model based on deep neural network, combining the CNN-GRU model, the vibration acceleration of wind turbines in different wake fields and yaw states are predicted, safety state analysis and full-field scheduling optimization are performed, and nonlinear optimization problems are used to solve nonlinear optimization problems and scheduling instructions are output.
It improves the accuracy and reliability of wind power farm power scheduling, ensures the safe state of the unit in the future time domain, and avoids the situation where scheduling instructions cannot be executed.
Smart Images
Figure CN120357555A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of wind power generation, and particularly to a reliable power dispatch method for a wind farm based on vibration acceleration prediction. Background Art
[0002] Currently, in order to obtain more wind energy and improve the power generation capacity within a limited area, the density between units in a wind farm is continuously increasing, and the scale of the units is continuously becoming larger. This leads to a significant wake effect in the wind farm, and requires coordinated dispatch of the whole field to ensure the power generation capacity of the whole field.
[0003] However, when the wind farm conducts power dispatch, it often directly issues power commands to the wind turbines. The wind turbines perform yaw control to achieve power dispatch. When such dispatch commands are issued, it is default that the wind turbines are in a safe state. Before or during the wind turbines' execution of yaw control, the safety state of the units will be judged. When in a non-safe state, the yaw control of the wind turbines will stop being executed. This results in the actual power dispatch of the wind farm not being able to be accurately completed.
[0004] Generally, the safety state during the yaw process of wind turbines mainly uses vibration acceleration as a monitoring index. Therefore, there is an urgent need for a reliable power dispatch method for a wind farm based on the vibration acceleration prediction of wind turbines. Summary of the Invention
[0005] The purpose of the present invention is to solve the problems existing in the prior art, and provide a reliable power dispatch method for a wind farm based on vibration acceleration prediction, which can effectively allocate the power generation of the units in the wind farm on the basis of accurately judging the safety state of the wind turbines at future times, avoid the situation where the power command cannot be executed after being issued, and can improve the accuracy and reliability of the power dispatch of the wind farm.
[0006] To achieve the above purpose, the present invention is realized through the following technical solutions:
[0007] A reliable power dispatch method for a wind farm based on vibration acceleration prediction, comprising the steps of:
[0008] S1. Based on a deep neural network, establish a vibration acceleration prediction model of a wind turbine under different wake fields and yaw states;
[0009] S2. Analyze the safety state of each unit in the wind farm to determine the unit combination participating in the dispatch;
[0010] S3. Determine the dynamic characteristic trajectory of the units participating in the power dispatch, and design a state trajectory model of the units participating in the power regulation;
[0011] S4. Design an overall dispatching optimization model for a wind power plant, including constructing the overall dispatching objective cost function of the wind farm, the safety constraint of the vibration acceleration of the units, and the overall wake dynamic model;
[0012] S5. Use an evolutionary algorithm to solve the corresponding non-linear optimal dispatching problem;
[0013] S6. If no feasible solution can be obtained, repeat steps S2 - S5; if a feasible solution is obtained, output the dispatching instruction.
[0014] Preferably, step S1 includes the steps of:
[0015] S11. Obtain historical data and simulation data, and divide various working conditions of the yaw process of the wind turbines, including wind speed, wind direction, wake field parameters, yaw angle, wind turbine torque of the unit, and unit speed;
[0016] S12. For various working conditions, use a deep neural network to establish a dynamic model of various driving factors and the vibration acceleration of the unit;
[0017] Among them, the dynamic model is:
[0018]
[0019] Among them, Acc ix (k) is the vibration acceleration of the i-th unit in the X direction, Acc iy (k) is the vibration acceleration of the i-th unit in the Y direction, W i (k) is the wake parameter of the impeller surface of the i-th unit, α i (k) is the axial induction factor of the i-th unit, τ i (k) represents the impeller torque of the i-th unit, γ i (k) is the control action vector composed of the yaw angles of the i-th unit, v i (k) is the input wind speed of the i-th unit under the wake field.
[0020] Preferably, in step S12, a deep neural network model is used for construction, combining the CNN model and the GRU to construct a combined CNN-GRU model, using CNN to mine the correlation between various factors and the vibration acceleration, and using GRU to mine the temporal characteristics of various factors in time.
[0021] Preferably, in step S2, a safety status analysis is carried out for the wind farm before dispatching. For the units with over-limit vibration acceleration, they do not participate in power dispatching.
[0022] Preferably, in step S3, based on a deep neural network with non-linearity, a state prediction trajectory model in the prediction time domain is established.
[0023] Preferably, in step S4, the full-field dispatching optimization model of the wind power plant is as follows:
[0024]
[0025] s.t.v i (k + 1|k) = v0(k)F i (k)
[0026]
[0027] γ min ≤γ i (k + h|k) ≤ γ max
[0028] v in ≤v i (k + h|k) ≤ v out
[0029] 0 ≤ p i (k + h|k) ≤ p e
[0030] 0 ≤ Acc ix (k + h|k) ≤ A eX
[0031] 0 ≤ Acc iy (k + h|k) ≤ A ey
[0032]
[0033] Among them, M is the prediction horizon length; Q > 0, R > 0 represent the weighting weights; v0(k + h - 1) is the input wind speed of the wind farm at a future time; F i (k) is the wake flow field model of the i-th unit; γ(k) is the control action vector composed of the yaw angles of the wind turbines; v(k) is the state vector composed of the input wind speeds of the wind turbines; p(k) is the output vector composed of the capturable power of the wind turbines; In addition, the input wind direction β0(k + h - 1) of the wind farm at a future time is also required; ρ(k) is the air density at time k, R is the impeller diameter of the wind turbine generator, γ i (k) is the yaw angle of the i-th unit; α i (k) is the axial induction factor of the i-th unit; τ i (k + h|k) is the impeller torque of the i-th unit; Acc ix (k + h|k) is the vibration acceleration of the i-th unit in the X direction; Acc iy (k + h|k) represents the vibration acceleration of the i-th unit in the Y direction; v i(k + h|k) is the input wind speed under the wake flow field of the i-th unit; τ i (k + h|k) is the wind turbine torque of the i-th unit; v in is the cut-in wind speed; v out is the cut-out wind speed; p e is the rated power of the wind turbine; A eX is the vibration acceleration safety threshold of the unit in the X direction, A ey is the vibration acceleration safety threshold of the unit in the Y direction; W i (k) is the wake parameter of the impeller surface of the i-th unit; G Acc (.) is the vibration acceleration prediction model of the wind turbine generator set.
[0034] Compared with the prior art, the beneficial effects of the present invention are as follows:
[0035] Aiming at the problem that the current wind farm scheduling process does not fully consider the yaw safety state of the units, the present invention proposes a reliable power scheduling method for wind farms based on the prediction of the vibration acceleration of the units, which can effectively determine the units participating in the power scheduling of the wind farm, accurately grasp the safety state of the units in the future time domain during the scheduling process, and improve the effectiveness and reliability of the power scheduling of the wind farm. Brief Description of the Drawings
[0036] Figure 1 is the neural network prediction model of the vibration acceleration of the unit;
[0037] Figure 2 is the schematic diagram of the safety judgment of the units participating in the wind farm;
[0038] Figure 3 is the flow chart of the reliable power scheduling of the wind farm. Detailed Embodiments
[0039] The present invention will be further described below in conjunction with specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and not to limit the scope of the present invention. In addition, it should be understood that after reading the content taught by the present invention, those skilled in the art can make various changes or modifications to the present invention, and these equivalent forms also fall within the scope defined by this application.
[0040] Embodiment: The present invention relates to a reliable power scheduling method for a wind farm based on the prediction of vibration acceleration, including the following steps:
[0041] I). Construction of the vibration acceleration prediction model of the wind turbine generator set
[0042] Establish a vibration acceleration prediction model of the wind turbine generator set under different wake flow fields and yaw states.
[0043] 1), Obtain historical data and simulation data, and divide various working conditions in the yaw process of the wind turbine generator set, including wind speed, wind direction, wake field parameters, yaw angle, wind turbine torque of the unit, and unit speed;
[0044] 2), For various working conditions, use a deep neural network to establish a dynamic model of various driving factors and the vibration acceleration of the unit.
[0045] The dynamic model is:
[0046]
[0047] Among them, Acc ix (k) is the vibration acceleration of the i-th unit in the X direction, Acc iy (k) is the vibration acceleration of the i-th unit in the Y direction, W i (k) is the wake parameter of the impeller surface of the i-th unit, α i (k) is the axial induction factor of the i-th unit, τ i (k) represents the impeller torque of the i-th unit, γ i (k) is the control action vector composed of the yaw angles of the i-th unit, v i (k) is the input wind speed of the i-th unit under the wake field.
[0048] Due to its non-linear characteristics and complex process, it is difficult to adopt the mechanism modeling process. Therefore, a deep neural network model is used for construction. Its network structure is as Figure 1 . Combine the CNN model with good feature extraction performance and GRU to construct a combined CNN-GRU model. Use CNN to mine the correlation between various factors and vibration acceleration, and use GRU to mine the time series characteristics of various factors in time, in order to comprehensively consider the time correlation and space correlation and improve the prediction accuracy of vibration acceleration.
[0049] II), Reliable power dispatch of wind farms based on vibration acceleration prediction
[0050] The dispatch flow chart is as Figure 3 shown, and specifically includes the following steps:
[0051] 1), As Figure 2 shown, to ensure reliable power dispatch, first conduct a safety status analysis on the wind farm before dispatch. For units with vibration acceleration exceeding the limit, they do not participate in power dispatch.
[0052] 2), Based on the non-linear deep neural network in (1), establish a state prediction trajectory model in the prediction time domain.
[0053] 3), Design the full-field scheduling objective cost function of the wind farm, and at the same time construct the safety constraint of the vibration acceleration of the unit;
[0054]
[0055] Among them, M is the prediction time domain length; Q>0, R>0 represent the weighting weights; v0(k+h-1) is the input wind speed of the wind farm at the future moment; F i (k) is the wake flow field model of the i-th unit; γ(k) is the control action vector composed of the yaw angles of the wind turbine; v(k) is the state vector composed of the input wind speeds of the wind turbine; p(k) is the output vector composed of the capturable power of the wind turbine; In addition, the input wind direction β0(k+h-1) of the wind farm at the future moment is also required; ρ(k) is the air density at the k-th moment, R is the impeller diameter of the wind power generation unit, γ i (k) is the yaw angle of the i-th unit; α i (k) is the axial induction factor of the i-th unit; τ i (k+h|k) is the impeller torque of the i-th unit; Acc ix (k+h|k) is the vibration acceleration of the i-th unit in the X direction; Acc iy (k+h|k) represents the vibration acceleration of the i-th unit in the Y direction; v i (k+h|k) is the input wind speed of the i-th unit under the wake flow field; τ i (k+h|k) is the wind turbine torque of the i-th unit; v in is the cut-in wind speed; v out is the cut-out wind speed; p e is the rated power of the wind turbine; A eX is the vibration acceleration safety threshold of the unit in the X direction, A ey is the vibration acceleration safety threshold of the unit in the Y direction; W i (k) is the wake flow parameter of the impeller surface of the i-th unit; G Acc (.) is the vibration acceleration prediction model of the wind power generation unit.
[0056] 4), Solve the corresponding non-linear optimization scheduling problem, and an intelligent evolutionary algorithm can be used to solve it.
[0057] 5), If a feasible solution cannot be obtained, then remove the units with excessive vibration, reconstruct the optimization problem twice, and solve it again; if a feasible solution is obtained, then output the scheduling instruction.
Claims
1. A reliable scheduling method for wind farm power based on vibration acceleration prediction, characterized in that Including the steps: S1. Based on a deep neural network, establish a vibration acceleration prediction model for a wind turbine generator under different wake fields and yaw states; S2. Conduct a safety status analysis of each unit in a wind farm and determine the participating unit combination for dispatching; S3. Determine the dynamic characteristic trajectory of the units participating in power dispatching and design a state trajectory model for the units participating in power regulation; S4. Design an overall dispatching optimization model for a wind power plant, including constructing an overall dispatching objective cost function for the wind power plant, a safety constraint for the vibration acceleration of the units, and an overall wake dynamic model; S5. Use an evolutionary algorithm to solve the corresponding non-linear optimization dispatching problem; S6. If no feasible solution can be obtained, repeat steps S2 - S5; if a feasible solution is obtained, output a dispatching instruction.
2. The reliable scheduling method for wind farm power based on vibration acceleration prediction according to claim 1, wherein, The said step S1 includes the steps: S11. Obtain historical data and simulation data, and divide various working conditions in the yaw process of the wind turbine generator, including wind speed, wind direction, wake field parameters, yaw angle, wind turbine torque of the unit, and unit speed; S12. For various working conditions, use a deep neural network to establish a dynamic model of various driving factors and the vibration acceleration of the unit; Among them, the dynamic model is: Among them, Acc ix (k) is the vibration acceleration of the i-th unit in the X direction, Acc iy (k) is the vibration acceleration of the i-th unit in the Y direction, W i (k) is the wake parameter of the impeller surface of the i-th unit, α i (k) is the axial induction factor of the i-th unit, τ i (k) characterizes the impeller torque of the i-th unit, γ i (k) is the control action vector composed of the yaw angles of the i-th unit, v i (k) is the input wind speed under the wake flow field of the i-th unit.
3. A reliable scheduling method for wind farm power based on vibration acceleration prediction according to claim 2, characterized in that In step S12, a deep neural network model is used for construction. The CNN model is combined with the GRU to construct a combined CNN - GRU model. The CNN is used to mine the correlation between various factors and the vibration acceleration, and the GRU is used to mine the temporal sequence characteristics of various factors in terms of time.
4. A reliable scheduling method for wind farm power based on vibration acceleration prediction according to claim 1, characterized in that In step S2, conduct a safety status analysis of the wind farm before dispatching. For the units with over-limit vibration acceleration, they do not participate in power dispatching.
5. A reliable scheduling method for wind farm power based on vibration acceleration prediction according to claim 1, characterized in that In step S3, based on a deep neural network with non-linearity, establish a state prediction trajectory model in the prediction time domain.
6. A reliable scheduling method for wind farm power based on vibration acceleration prediction according to claim 1, characterized in that In step S4, the overall dispatching optimization model for the wind power plant is as follows: s.t.v i (k + 1|k) = v0(k)F i (k) γ min ≤γ i (k + h|k) ≤ γ max v in ≤v i (k + h|k) ≤ v out 0 ≤ p i (k + h|k) ≤ p e 0 ≤ Acc ix (k + h|k) ≤ A eX 0 ≤ Acc iy (k + h|k) ≤ A ey where M is the prediction horizon length; Q>0, R>0 represent the weighting weights; v0(k+h-1) is the input wind speed at the future time of the wind farm; F i (k) is the wake flow field model of the i-th unit; γ(k) is the control action vector composed of the yaw angles of the wind turbine; v(k) is the state vector composed of the input wind speeds of the wind turbine; p(k) is the output vector composed of the capturable power of the wind turbine; in addition, the input wind direction β0(k+h-1) at the future time of the wind farm is also required; ρ(k) is the air density at time k, R is the impeller diameter of the wind turbine generator set, γ i (k) is the yaw angle of the i-th unit; α i (k) is the axial induction factor of the i-th unit; τ i (k+h|k) is the impeller torque of the i-th unit; Acc ix (k+h|k) is the vibration acceleration of the i-th unit in the X direction; Acc iy (k+h|k) represents the vibration acceleration of the i-th unit in the Y direction; v i (k+h|k) is the input wind speed of the i-th unit under the wake flow field; τ i (k+h|k) is the wind turbine torque of the i-th unit; v in is the cut-in wind speed; v out is the cut-out wind speed; p e is the rated power of the wind turbine; A eX is the safety threshold of the vibration acceleration of the unit in the X direction, A ey is the safety threshold of the vibration acceleration of the unit in the Y direction; W i (k) is the wake parameter of the impeller surface of the i-th unit; G Acc (.) is the vibration acceleration prediction model of the wind turbine generator set.