Markov-based self-triggering yaw control method

Abstract

The invention relates to a self-triggering yaw control method based on semi-Markov. The self-triggering yaw control method comprises the following steps that S1, the real-time yaw angle of a wind turbine generator and the yaw angle of the last control moment are obtained; s2, constructing an optimal control strategy based on a Markov decision process; S3, determining a control opportunity by adopting an event triggering mechanism; and S4, controlling input to act on a yaw driving system to realize a yaw action of the wind turbine generator. The system comprises a master control PLC module with a built-in Markov event-triggered yaw control algorithm, a wind speed and direction sensor, a cabin position sensor, a yaw frequency converter, a yaw motor, a yaw speed reducer, a yaw gear ring and a hydraulic braking system. Unnecessary triggering of the controller is reduced through an event triggering mechanism, mechanical abrasion is reduced, meanwhile, an optimal control strategy is generated through the Markov decision process, the system response speed and control precision are improved, accurate wind alignment is achieved, and the wind energy utilization efficiency is improved.

Description

Technical Field

[0001] The present invention relates to a self-triggered yaw control method based on Markov and belongs to the technical field of wind power generation. Background Art

[0002] With the continuous growth of the global demand for clean energy, wind power generation, as an important form of renewable energy utilization, has developed rapidly. The scale of wind farms is expanding day by day, and numerous wind turbines are densely distributed within a limited area. During the operation of a wind farm, the accurate wind alignment control of wind turbines has an important impact on power generation efficiency.

[0003] In the current research on wind turbine yaw control, traditional methods such as PID and fuzzy control are mainly used. These methods have the following problems: First, frequent triggering of the controller leads to large mechanical losses in the yaw system; second, the control strategy is relatively fixed and difficult to adapt to complex and variable wind conditions; finally, the control response speed is slow and it is difficult to achieve accurate wind alignment. Traditional yaw control strategies usually perform control at fixed time intervals. This method not only increases the mechanical wear of the system but also makes it difficult to balance the relationship between control accuracy and energy consumption.

[0004] In the prior art, although there are also solutions that use intelligent control algorithms to optimize yaw control, most of them focus on the optimization of a single control target and fail to well solve the contradiction between control frequency and control accuracy. In addition, these methods often require frequent adjustment of the yaw angle, resulting in frequent start and stop of the yaw motor, which not only increases energy consumption but also reduces the service life of the equipment. Therefore, how to reduce the frequency of control actions while ensuring wind alignment accuracy and improve the overall efficiency of the system has become a technical problem to be solved urgently. With the continuous growth of the global demand for clean energy, wind power generation, as an important form of renewable energy utilization, has developed rapidly. The scale of wind farms is expanding day by day, and numerous wind turbines are densely distributed within a limited area. During the operation of a wind farm, the wake interaction between wind turbines has become a significant problem. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a self-triggered yaw control method based on semi-Markov, which realizes accurate wind alignment and reduces mechanical losses by optimizing the control strategy and triggering mechanism.

[0006] The present invention adopts the following technical solutions:

[0007] A self-triggered yaw control method based on semi-Markov of the present invention adopts the following steps:

[0008] Step S1, obtain the real-time yaw angle of the wind turbine and the yaw angle at the previous control moment as the input quantity of the control system;

[0009] Step S2: Process the set target yaw angle through a control strategy based on the Markov decision process: establish a yaw angle state space and a rudder angle action space, design a state transition probability matrix and a reward function, and solve the optimal control strategy through the value iteration algorithm;

[0010] Step S3: Adopt an event-triggered mechanism to determine the control timing: calculate the deviation between the current yaw angle and the yaw angle at the last control moment, trigger a control action when the deviation exceeds the set threshold, and generate the corresponding control input according to the optimal strategy;

[0011] Step S4: Apply the control input to the yaw drive system to achieve the yaw action of the wind turbine and complete the wind regulation.

[0012] In step S1 of the present invention, obtaining the operating state of the wind turbine and initializing the control system mainly includes obtaining the operating parameters of the wind turbine, constructing the state space, constructing the control action space, and initializing the system parameters;

[0013] First, collect the yaw angle θcurrent of the wind turbine from the nacelle position sensor in real time, and at the same time read the yaw angle θlast at the last control moment from the system record, and obtain the current power, speed and other operating parameters of the wind turbine; second, establish the state space of the yaw control system, select the yaw angle range from -30 degrees to 30 degrees, and discretize it at intervals of 0.5 degrees to form a discrete state space containing 121 state points; third, construct the control action space, select the rudder angle range from -10 degrees to 10 degrees, and discretize it at intervals of 1 degree to form a discrete action space containing 21 action points. Fourth, complete the initialization of the system parameters, including setting the initial value of the event trigger threshold δ, determining the weight coefficient of the reward function, and setting the discount factor γ, laying a foundation for the subsequent optimization of the control strategy and the implementation of the event-triggered mechanism; this initialization process fully considers the physical characteristics and control requirements of the yaw system of the wind turbine, and through reasonable parameter settings and state space construction, ensures that the control system can effectively achieve the control objectives of accurate wind alignment and reduction of mechanical losses.

[0014] In step S2 of the present invention, the yaw angle state space adopts a discrete state space from -30 degrees to 30 degrees with an interval of 0.5 degrees; the rudder angle action space adopts a discrete rudder angle space from -10 degrees to 10 degrees with an interval of 1 degree. The state transition probability matrix describes the probability of transitioning to the next state after taking a certain action in the current state, and its calculation expression is:

[0015] P(s′|s,a) = 0.8 when s' is the main target state;

[0016] P(s′|s,a) = 0.1 when s' is an adjacent state;

[0017] P(s′|s,a) = 0, for other cases;

[0018] where s is the current yaw angle state, a is the rudder angle control action, and s' is the next yaw angle state;

[0019] The design of the reward function in step S2 of the present invention includes two parts: state reward and action reward. The state reward is inversely proportional to the yaw angle error, and the action reward is inversely proportional to the amplitude of the control action. The specific expression is as follows:

[0020] R(s,a,s′) = -5|θ| - 0.1|a|

[0021] where θ is the yaw angle value, a is the rudder angle control amount, the negative sign represents the penalty term, and the coefficients 5 and 0.1 are the weight coefficients of the state error and the control action respectively. By adjusting these two coefficients, the relationship between control precision and control frequency can be balanced.

[0022] The design of the event trigger mechanism in step S3 of the present invention adopts a trigger strategy based on the deviation threshold, and its trigger condition expression is:

[0023] |θcurrent - θlast| > σ

[0024] where θcurrent is the current yaw angle, θlast is the yaw angle at the last control moment, and σ is the trigger threshold (set to 0.5 degrees). When the deviation exceeds the threshold, the control action is triggered, thereby reducing unnecessary control operations;

[0025] First, calculate the deviation value |θcurrent - θlast| between the current yaw angle θcurrent and the yaw angle θlast at the last control moment in real time, and compare this deviation value with the preset trigger threshold σ. When the deviation value exceeds the trigger threshold, the control event is triggered, and the system searches for the corresponding control action from the optimal strategy according to the current state; when the deviation value does not exceed the trigger threshold, the system maintains the current state and does not generate new control actions. For the triggered control event, the system will generate the corresponding rudder angle control input according to the optimal strategy obtained from the Markov decision process and in combination with the current yaw angle state. This event trigger mechanism avoids the frequent adjustment problem brought by the traditional fixed time interval control method by dynamically judging the control demand, effectively reducing mechanical wear; at the same time, the setting of the trigger threshold fully considers the control precision and the system response characteristics, ensuring timely control at necessary moments and avoiding energy waste caused by over-control, achieving a good balance between control efficiency and system reliability;

[0026] In step S2 of the present invention, the optimal control strategy is solved by the value iteration algorithm, and its iterative process includes: initializing the value function, calculating the action value, updating the value function, and repeating the iteration until convergence. The value function update formula is:

[0027] V(s) = max[ΣP(s′|s,a)(R(s,a,s′) + γV(s′)]

[0028] where γ is the discount factor (with a value of 0.95).

[0029] In step S4 of the present invention, the execution of the control input adopts a variable - frequency speed - regulation method. According to the rudder - angle action value determined by the optimal strategy, the speed of the yaw motor is adjusted through a frequency converter to achieve a smooth yaw action.

[0030] The beneficial effects of the present invention are as follows:

[0031] 1. The trigger times of the controller are significantly reduced through the event - trigger mechanism, reducing mechanical losses;

[0032] 2. The optimal control strategy based on the Markov decision process improves the wind - facing accuracy;

[0033] 3. The self - adaptive adjustment of control parameters enhances the robustness of the system;

[0034] 4. The variable - frequency speed - regulation realizes smooth control and extends the service life of the equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Att Figure 1 is the flowchart of the method of the present invention;

[0036] Att Figure 2 is the historical yaw - angle diagram;

[0037] Att Figure 3 is the historical control diagram;

[0038] Att Figure 4 is the historical event - trigger diagram; DETAILED DESCRIPTION OF THE INVENTION

[0039] The present invention will be described in detail below with reference to the accompanying drawings.

[0040] As shown in Att Figure 1 The present invention is a self - triggered yaw control method based on semi - Markov, which is characterized by adopting the following steps:

[0041] Step S1: Regarding the acquisition of the wind - turbine generator set state and system initialization, select the 5MW unit of the National Renewable Energy Laboratory (NREL) of the United States as the research object. The hub height is 90m, the wind - wheel diameter D is 126m. The system state space selects the yaw - angle range from - 30 degrees to 30 degrees, and is discretized at intervals of 0.5 degrees to form a discrete state space containing 121 state points. The control action space selects the rudder - angle range from - 10 degrees to 10 degrees, and is discretized at intervals of 1 degree to form a discrete action space containing 21 action points;

[0042] According to the characteristics of the Markov decision process, the state transition probability matrix describes the probability of transitioning to the next state after taking a certain action in the current state, and its calculation expression is as follows:

[0043] P(s′|s,a) = 0.8 when s' is the main target state;

[0044] P(s′|s,a) = 0.1 when s' is an adjacent state;

[0045] P(s′|s,a) = 0 in other cases;

[0046] Among them, s is the current yaw angle state, a is the rudder angle control action, and s' is the next yaw angle state. The design of the reward function includes two parts: state reward and action reward, and its expression is:

[0047] R(s,a,s′) = -5|θ| - 0.1|a|

[0048] Among them, θ is the yaw angle value, a is the rudder angle control amount, the negative sign represents the penalty term, and the coefficients 5 and 0.1 are the weight coefficients of the state error and the control action respectively. By adjusting these two coefficients, the relationship between control accuracy and control frequency can be balanced;

[0049] In step S2, the control strategy optimization based on the Markov decision process uses the value iteration algorithm. Initialize the value function V(s) as a zero vector, and then iteratively update the value of each state. For each state s, calculate the expected return Q(s,a) after taking different actions a in this state:

[0050] Q(s,a) = ∑P(s′|s,a)[R(s,a,s′) + γV(s′)]

[0051] Among them, γ is the discount factor (taking the value of 0.95), which represents the degree of emphasis on future returns. Then update the state value:

[0052] V(s) = max[Q(s,a)]

[0053] Repeat the above process until the value function converges, that is, the maximum difference between two consecutive iterations is less than a preset threshold (such as 1e-6). The final optimal strategy is the action sequence that maximizes Q(s,a);

[0054] To verify the convergence of the algorithm, different initial states were tested. The results show that when the discount factor γ = 0.95, the algorithm usually converges to a stable solution within 500 - 1000 iterations; the convergence process of the value function is as Figure 2As shown, the horizontal axis represents the number of iterations, and the vertical axis represents the maximum change in the value function. It can be seen from the figure that as the number of iterations increases, the change in the value function gradually decreases and finally stabilizes.

[0055] The event-triggered control mechanism in step S3 adopts a design scheme based on a deviation threshold. The system uses a nacelle position sensor to collect the yaw angle θcurrent of the wind turbine in real time and compares it with the yaw angle θlast at the previous control moment. The event-triggered judgment condition is:

[0056] |θcurrent - θlast| > σ

[0057] where σ is the trigger threshold (initially set to 0.5 degrees). When the deviation exceeds the threshold, the system triggers a control event and updates the control action; otherwise, it maintains the current control output unchanged. This mechanism significantly reduces the trigger frequency of the controller and reduces mechanical wear.

[0058] To verify the effectiveness of the event-triggered mechanism, a comparative experiment was conducted. The experiment used traditional fixed-time-interval control and the event-triggered control of the present invention respectively, and ran for 100 time steps under the same working conditions. The results showed that the event-triggered control scheme reduced the average number of control actions by about 70% while maintaining similar control accuracy. The experimental results are as Figure 3 shown, where the blue line represents the yaw angle change curve, the red line represents the control action curve, and the black pulse represents the trigger event.

[0059] By adjusting the size of the trigger threshold σ, the balance between the control frequency and the control accuracy can be achieved. A larger threshold can further reduce the trigger times but may affect the control accuracy, while a smaller threshold can improve the control accuracy but increase the trigger frequency. In practical applications, the appropriate threshold can be set according to specific requirements.

[0060] The control execution in step S4 adopts a variable-frequency speed regulation method. According to the rudder angle control amount calculated by the optimal strategy, the speed of the yaw motor is adjusted by the frequency converter to achieve smooth control. The control system has an adaptive ability during the execution process and can automatically adjust the control parameters according to the actual operation effect. Among them, the adaptive adjustment strategy of the trigger threshold δ is:

[0061] σnew = σold * (1 + k * Δe)

[0062] where Δe is the control error change rate and k is the adjustment coefficient (with a value of 0.1). When the control effect is good, the trigger threshold is appropriately increased to reduce the control frequency; when the control error increases, the trigger threshold is decreased to improve the control accuracy.

[0063]

[0064]

[0065] To verify the effectiveness of the control method of the present invention, simulation tests were carried out on the NREL 5MW wind turbine. The test conditions were as follows: the initial yaw angle was 15 degrees, the target yaw angle was 0 degrees, and the running time was 100 steps; the simulation results are as Figure 4 shown. Compared with the traditional PID control, the method of the present invention has the following advantages:

[0066] 1. The control accuracy is improved by about 30%, and the steady-state yaw angle error is controlled within the range of ±0.5 degrees;

[0067] 2. The number of control actions is reduced by about 70%, significantly reducing mechanical wear;

[0068] 3. The system response time is shortened by about 20%, and the dynamic performance is improved;

[0069] 4. The variable frequency speed regulation realizes smooth control, and the transition process is more stable;

[0070] In summary, the present invention optimizes the control strategy through the Markov decision process, combines the event-triggered mechanism to reduce the control frequency, realizes accurate wind alignment control while reducing mechanical losses. During the actual operation process, the system continuously optimizes the control strategy through online learning, records the state transition of each control action and the obtained rewards, regularly updates the state transition probability matrix and recalculates the optimal control strategy. At the same time, the control system has an adaptive ability, can automatically adjust the control parameters according to the actual operation conditions, automatically adjust the trigger threshold δ according to the control effect, adjust the reward function weight coefficient according to the system response characteristics. In addition, it can also adjust the resolution of the state space and action space according to the control accuracy requirements. Such a design realizes the event-triggered adaptive optimal control, which not only ensures the control accuracy but also reduces mechanical losses. This method has good engineering practicability and can be popularized and applied to the yaw control system of large wind farms.

[0071] A yaw control method for a wind turbine based on Markov event triggering of the present invention, its installation method, connection method or setting method are all common methods, and any method that can achieve its beneficial effects can be implemented; the system includes a main control PLC module with a built-in Markov event-triggered yaw control algorithm, a wind speed and direction sensor, a nacelle position sensor, a yaw frequency converter, a yaw motor, a yaw reducer, a yaw ring gear and a hydraulic braking system.

[0072] Obviously, the above embodiments are only certain embodiments of the present invention and are not used to limit the present invention; at the same time, for those skilled in the art, the present invention can have various changes and modifications; any modifications, equivalent replacements and improvements made within the spirit and scope of the present invention are all within the protection scope of the present invention.

Claims

1. A semi-Markov based self-triggered yaw control method, characterized in that: Use the following steps: Step S1, obtaining the real-time yaw angle of the wind turbine and the yaw angle at the last control moment as input quantities of the control system; Step S2, processing the set target yaw angle through a control strategy based on the Markov decision process: establishing a yaw angle state space and a rudder angle action space, designing a state transition probability matrix and a reward function, and solving the optimal control strategy through a value iteration algorithm; Step S3, using an event trigger mechanism to determine the control timing: calculating the deviation between the current yaw angle and the yaw angle at the last control moment, triggering the control action when the deviation exceeds the set threshold, and generating corresponding control input according to the optimal strategy; Step S4: Control input acts on the yaw drive system to achieve the yaw action of the wind turbine and complete wind regulation.

2. The semi-Markov based self-triggered yaw control method according to claim 1, characterized in that: In step S2, the yaw angle state space adopts a discrete state space of -30 degrees to 30 degrees with an interval of 0.5 degrees; the rudder angle action space adopts a discrete rudder angle space of -10 degrees to 10 degrees with an interval of 1 degree. The state transition probability matrix describes the probability of transitioning to the next state after taking a certain action in the current state, and the calculation formula is as follows: P(s′|s,a)=0.8, when s′ is the main target state; P(s′|s,a)=0.1, when s′ is an adjacent state; P(s′|s,a)=0,other cases; Among them, s is the current yaw angle state, a is the rudder angle control action, and s′ is the next yaw angle state.

3. The semi-Markov based self-triggered yaw control method according to claim 2, characterized in that: The reward function design in step S2 includes state reward and action reward. The state reward is inversely proportional to the yaw angle error and is used to ensure wind accuracy. The action reward is inversely proportional to the control action amplitude and is used to reduce the control frequency. The specific expression is as follows: R(s,a,s′)=-5|θ|-0.1|a| Among them, θ is the yaw angle value, a is the rudder angle control amount, the negative sign represents the penalty term, and the coefficients 5 and 0.1 are the weight coefficients of the state error and the control action respectively. By adjusting these two coefficients, the relationship between control accuracy and control frequency can be balanced.

4. The semi-Markov based self-triggered yaw control method according to claim 1, characterized in that: The design of the event trigger mechanism in step S3 adopts a trigger strategy based on a deviation threshold, and its trigger condition expression is as follows: |θcurrent-θlast|>σ| Among them, θcurrent is the current yaw angle, θlast is the yaw angle at the last control moment, and σ is the trigger threshold (set to 0.5 degrees). When the deviation exceeds the threshold, the control action is triggered, thereby reducing unnecessary control operations.

5. The semi-Markov based self-triggered yaw control method according to claim 1 or 2, characterized in that: In step S2, the optimal control strategy is solved by a value iteration algorithm, and the iteration process is as follows: 1) Initialize the value function V(s) = 0; 2) For each state s, calculate the action value: Q(s,a)=ΣP(s′|s,a)[R(s,a,s′)+γV(s′)] Where γ is the discount factor (value is 0.95); 3) Update the value function: V(s)=max[Q(s,a)] 4) Repeat steps 2-3 until convergence.

6. The semi-Markov based self-triggered yaw control method according to claim 1, characterized in that: The execution of the control input in step S4 adopts the variable frequency speed regulation mode. According to the rudder angle action value determined by the optimal strategy, the yaw motor speed is adjusted by the frequency converter to achieve smooth yaw action and reduce mechanical shock.

Images