A power generation quality intelligent control method of a tidal current energy power generation system

Abstract

The application discloses a kind of tidal current energy power generation system's power generation quality intelligent control method. The deep deterministic policy gradient algorithm based on optimization learning parameter is constructed, the state space of tidal current energy power generation system is handled by algorithm and the network parameter of algorithm is updated, and the system action is obtained after processing;Determine the termination state of current state space, and real-time power generation quality intelligent control is carried out on system action by optimizing learning parameter. The method of the present application realizes the real-time optimization control of the power generation voltage and frequency of tidal current energy power generation system through deep reinforcement learning, can effectively guide the learning process, adjust the convergence to the optimal control strategy in real time, and be applied to the actual tidal current energy power generation system to adapt to the change of tidal current velocity, adapt to complex marine environment and tidal current change, enhance the robustness of the system, ensure that the output power generation voltage and frequency meet the requirements of power grid, ensure the stability of power generation quality in dynamic environment and improve overall performance.

Description

TECHNICAL FIELD

[0001] The application relates to an intelligent control method for power generation quality, and relates to deep reinforcement learning, in particular to an intelligent control method for power generation quality of a tidal current power generation system. BACKGROUND

[0002] With the increasing demand for renewable energy worldwide, tidal current energy, as a clean and renewable energy resource, has attracted widespread attention. The tidal current power generation system utilizes the kinetic energy of ocean tides to convert into electric energy, and has the advantages of high efficiency and low pollution. However, one of the major challenges in the tidal current power generation process is the stability of power generation quality, mainly including the control of power generation voltage and frequency. Under the influence of factors such as tidal velocity change and load fluctuation, the output of the generator may be unstable, leading to a decrease in power grid quality and affecting the safe operation of equipment.

[0003] Traditional control methods such as PID control often face the problems of slow adjustment speed and poor adaptability when dealing with nonlinear and dynamically changing tidal current power generation systems, and it is difficult to meet the strict requirements of modern power grids for power generation quality. Therefore, a new control method is needed to improve the power generation quality and system stability of tidal current power generation. SUMMARY

[0004] In order to solve the problems in the background art, the application provides an intelligent control method for power generation quality of a tidal current power generation system. The method uses deep reinforcement learning technology to realize real-time optimization control of the power generation voltage and frequency of the tidal current power generation system through online learning of the control strategy.

[0005] The technical scheme adopted by the application is:

[0006] The intelligent control method for power generation quality of the tidal current power generation system of the application comprises:

[0007] Step 1) constructing a deep deterministic policy gradient DDPG (Deep Deterministic Policy Gradient) algorithm based on optimized learning parameters.

[0008] Step 2) constructing the power generation voltage, frequency and power of the tidal current power generation system as a state space and processing and updating the network parameters of the deep deterministic policy gradient DDPG algorithm through the deep deterministic policy gradient DDPG algorithm, and obtaining the action of the tidal current power generation system after processing.

[0009] Step 3) Determine whether the state space of the current tidal energy power generation system has reached the terminal state. If not, continue with the next round of learning, input the state space of the tidal energy power generation system into the deep deterministic policy gradient DDPG algorithm after the network parameters are updated, and repeat steps 2)-3) until the state space of the tidal energy power generation system reaches the terminal state. Save the network parameters and weights of the current deep deterministic policy gradient DDPG algorithm, and control the tidal energy power generation system in real time through the actions obtained by the current deep deterministic policy gradient DDPG algorithm. Then, in real time, optimize the learning parameters to intelligently control the power generation quality of the tidal energy power generation system.

[0010] In step 1), the optimization of the learning parameter P is as follows:

[0011] P=w1·MAE v +w2·MAE f +w3·MAE p +w4·Std v +w5·Std f +w6·Std p +w7·Δr

[0012] Δr=r t -r t-n

[0013] Among them, w1 and w4 are the mean absolute error MAE of the generated voltage v of the tidal energy generation system v and standard deviation Std v The weight coefficients of w2 and w5 are the mean absolute error MAE of the power generation frequency f of the tidal energy generation system. f and standard deviation Std f The weight coefficients of w3 and w6 are the mean absolute error MAE of the power generation power p of the tidal energy generation system. p and standard deviation Std p The weight coefficient of the generated voltage v is the standard deviation Std v Measures the fluctuation of voltage output, the standard deviation Std of the power generation frequency f f Measures the fluctuation of frequency output, standard deviation Std of power generation p p Measures the volatility of power output; Δr and w7 are the reward r at time tn of the deep deterministic policy gradient DDPG algorithm within n time steps. t-n and the reward r at the current time t t The reward change between , in the specific implementation, n = 100 time steps, w7 are the weight coefficients of the reward change Δr; the weight coefficients are adjusted according to the importance of each indicator in the control task.

[0014] The reward r of the current t moment of the deep deterministic policy gradient DDPG algorithm t The specific implementation is as follows:

[0015] r t =-(w v ·|v t -v target |+w f ·|f t -f target |+w p ·|p t -p target |)

[0016] Wherein, v t and v target are the power generation voltage and its target value of the tidal current power generation system at the current t moment, f t and f target are the power generation frequency and its target value of the tidal current power generation system at the current t moment, p t and p target are the power generation power and its target value of the tidal current power generation system at the current t moment, w v , w f and w p are weight coefficients of the power generation voltage v, frequency f and power p of the tidal current power generation system, used for balancing the influence of voltage, frequency and power on the total reward, and these weights can be adjusted according to the actual needs of the application scene, and the greater the weight represents the main share of the parameter in the control target.

[0017] In the step 1), the deep deterministic policy gradient DDPG algorithm includes an action current Actor-online network μ, an action target Actor-target network μ', a Q value current Critic-online network Q, a Q value target Critic-target network Q' and an experience replay pool, and the four network structures use deep multi-layer perceptron DNN (Deep Neural Networks), which has multiple hidden layers and has good learning and fitting capabilities, and can better adapt to the complex and variable nonlinear tidal current power generation system; the action current Actor-online network μ performs iterative update of its network parameters θ μ , and generates a current action a t according to the state space s t of the tidal current power generation system, to interact with the environment of the tidal current power generation system to generate a next state space s t+1 and a current reward r t ; the action target Actor-target network μ' generates a next space state s t+1 according to the experience replay pool.generating an optimal next action a t+1 and periodically copying the network parameters θ of the action current Actor-online network μ μ to update its own network parameters θ μ′ ; the Q value current Critic-online network Q obtains its own network parameters θ Q and obtains the current Q value Q(s t , a t | θ Q ); the Q value target Critic-target network Q' obtains the target Q value Q'(s i , μ'(s i | θ μ′ ) | θ Q′ ), s i represents the state space obtained in the i-th iteration, and periodically copies the network parameters θ of the Q value current Critic-online network Q Q to update its own network parameters θ Q′ .

[0018] In step 2), the power generation voltage, frequency and power of the tidal current power generation system are constructed into a state space and processed by the deep deterministic policy gradient DDPG algorithm, and the network parameters of the deep deterministic policy gradient DDPG algorithm are updated, specifically as follows:

[0019] Step 2.1) uses a random initialization method to initialize the network parameters θ of the current Actor-online network μ of the deep deterministic policy gradient DDPG algorithm μ and the network parameters θ of the Q value current Critic-online network Q Q and the experience replay pool D, and copies the initialized network parameters θ of the current Actor-online network μ μ to the network parameters θ of the action target Actor-target network μ' μ′ , and copies the initialized network parameters θ of the Q value current Critic-online network Q Q to the network parameters θ of the Q value target Critic-target network Q' Q′ .

[0020] Step 2.2) initializes the random noise E of the deep deterministic policy gradient DDPG algorithm, and simultaneously obtains the current state space s t of the tidal current power generation system, obtains the current action a t based on the initialized random noise E and the current state space s t , a t = μ(s t | θμ )+E.

[0021] Step 2.3) the tidal current energy generation system performs a current action a t , obtains a current reward r t and a next state space s t+1 , the current state space s t of the tidal current energy generation system, the current action a t , the current reward r t and the next state space are jointly constructed into experience data (s t , a t , r t , s t+1 ) and stored in an experience replay pool D, the current state space s t is updated to the next state space s t+1 .

[0022] Step 2.4) M experience data are randomly sampled from the experience replay pool D, and a target value y is calculated, a Q value of a Q value current Critic-online network Q is calculated according to the target value y, and network parameters θ Q of the Q value current Critic-online network Q are updated using a minimization loss function L to update network parameters θ Q of the current Actor-online network μ according to the updated network parameters θ μ of the Q value current Critic-online network Q.

[0023] Step 2.5) an action target Actor-target network μ' copies network parameters θ μ of an action current Actor-online network μ to update its own network parameters θ μ′ ; a Q value target Critic-target network Q' copies network parameters θ Q of a Q value current Critic-online network Q to update its own network parameters θ Q′ .

[0024] Step 2.6) after the update, an action of the tidal current energy generation system is obtained, including the amplitude, frequency and phase of the excitation current of the inverter controller (5) of the tidal current energy generation system.

[0025] The method of the present invention utilizes deep reinforcement learning technology to achieve real-time optimal control of the voltage, frequency, and power of a tidal power generation system through online learning of control strategies. Using the actor-critic structure of the deep deterministic policy gradient (DDPG) algorithm, the intelligent agent learns by observing the system state, the actions taken, and the designed reward function. The actor network generates control actions to optimize the voltage, frequency, and power of the tidal power generation system, and the critic network evaluates the actor network based on the system's Q-value. An experience replay mechanism is introduced, whereby empirical data samples generated by the actor network's interaction with the environment are stored in an experience replay pool. Batch data samples are then extracted for online learning to update network parameters, thereby generating the optimal control strategy.

[0026] In the step 3), the power generation quality of the tidal energy power generation system is intelligently controlled by optimizing the learning parameters in real time. Specifically, when the optimized learning parameter P exceeds the preset parameter threshold P threshold When , re-evaluation and training are performed, the current state space of the tidal energy generation system is input into the deep deterministic policy gradient DDPG algorithm after the network parameters are updated and steps 2)-3) are repeated until the optimized learning parameter P does not exceed the preset parameter threshold P threshold , thereby intelligently controlling the power generation quality of the tidal energy power generation system through current actions.

[0027] The beneficial effects of the present invention are:

[0028] The method of the present invention realizes real-time optimization control of the generation voltage and frequency of the tidal energy power generation system through deep reinforcement learning, can effectively guide the learning process, adjust and converge to the optimal control strategy in real time, and is applied to actual tidal energy power generation systems. It can handle the nonlinear and strong interference characteristics of the tidal energy power generation system to adapt to changes in tidal flow velocity, adapt to complex marine environments and tidal changes, enhance the robustness of the system, ensure that the output generation voltage and frequency meet the requirements of the power grid, ensure the stability of power generation quality in a dynamic environment and improve the overall performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 is a control flow chart of the method of the present invention;

[0030] Figure 2 This is a working principle diagram of the doubly-fed induction generator of the present invention;

[0031] In the figure: 1. Gearbox, 2. Doubly-fed induction generator, 3. Bidirectional frequency converter, 4. Power grid, 5. Inverter controller, 6. Operation status detector. DETAILED DESCRIPTION

[0032] In order to make the purposes, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described in detail below with reference to the drawings of the present application. It should be understood that the specific embodiments described herein are only used to explain the present application and not used to limit the present application. In addition, the technical features involved in each embodiment of the present application described below can be combined with each other as long as they do not conflict with each other.

[0033] As shown in Figure 1 , the power generation quality intelligent control method of the tidal power generation system of the present application is as follows:

[0034] Step 1) Construct a deep deterministic policy gradient DDPG algorithm based on optimized learning parameters, and the optimized learning parameters P are as follows:

[0035] P=w1·MAE v +w2·MAE f +w3·MAE p +w4·Std v +w5·Std f +w6·Std p +w7·Δr

[0036] Δr=r t -r t-n

[0037] Wherein, w1 and w4 are weight coefficients of the average absolute error MAE v and the standard deviation Std v of the power generation voltage v of the tidal power generation system, w2 and w5 are weight coefficients of the average absolute error MAE f and the standard deviation Std f of the power generation frequency f of the tidal power generation system, w3 and w6 are weight coefficients of the average absolute error MAE p and the standard deviation Std p of the power generation power p of the tidal power generation system, the standard deviation Std v of the power generation voltage v measures the volatility of the voltage output, the standard deviation Std f of the power generation frequency f measures the volatility of the frequency output, and the standard deviation Std p of the power generation power p measures the volatility of the power output; Δr and w7 are respectively the reward change amount between the reward r t-n at t-n time of the deep deterministic policy gradient DDPG algorithm in n time steps and the reward r t at the current t time, and n=100 time steps in specific implementation, and w7 is the weight coefficient of the reward change amount Δr; the weight coefficients are adjusted according to the importance of each index in the control task.

[0038] The mean absolute error (MAE) of the generated voltage v of the tidal energy generation system v The details are as follows:

[0039]

[0040] Where N is the total time step, v t and v target are the power generation voltage and target value of the tidal energy generation system at the current moment t.

[0041] The mean absolute error (MAE) of the generation frequency f of the tidal energy generation system f and the mean absolute error MAE of the generated power p p The calculation method is similar to the mean absolute error MAE of the generated voltage v v .

[0042] The present invention designs a deep deterministic policy gradient (DDPG) algorithm with optimized learning parameters P and a timed evaluation of the actions of the tidal energy generation system to determine a strategy for continued learning. The strategy combines the tracking errors of voltage, frequency and power, system stability and changes in reward signals.

[0043] The reward r at the current time t of the deep deterministic policy gradient DDPG algorithm t The details are as follows:

[0044] r t =-(w v ·|v t -v target |+w f ·|f t -f target |+w p ·|p t -p target |)

[0045] Among them, v t and v target are the power generation voltage and target value of the tidal energy generation system at the current time t, f t and f target are the power generation frequency and target value of the tidal energy power generation system at the current time t, p t and p target are the power generation power and target value of the tidal energy generation system at the current moment t, w v 、w f and w pThese are the weight coefficients of the generated voltage v, frequency f, and power p of the tidal energy generation system, which are used to balance the impact of voltage, frequency, and power on the total reward. These weights can be adjusted according to the actual needs of the application scenario. The larger the weight, the more important the parameter is in the control target.

[0046] The deep deterministic policy gradient DDPG algorithm includes the action current Actor-online network μ, the action target Actor-target network μ′, the Q value current Critic-online network Q, the Q value target Critic-target network Q′ and the experience replay pool. These four network structures use the deep multi-layer perceptron DNN (Deep Neural Networks), which has multiple hidden layers and has good learning and fitting capabilities. It can better adapt to complex and changeable nonlinear tidal power generation systems; the action current Actor-online network μ performs its own network parameter θ μ Iterative update, and according to the state space s of the tidal energy generation system t Generate current action a t , to interact with the environment of the tidal energy generation system to generate the next state space s t+1 and the current reward r t ; The action target Actor-target network μ′ replays the next spatial state s sampled from the experience pool t+1 Generate the optimal next action a t+1 , and periodically copy the network parameters θ of the current Actor-online network μ μ To update its own network parameters θ μ′ ; Q value current Critic-online network Q performs its own network parameters θ Q Iterative update and obtain the current Q(s t ,a t |θ Q ) value; Q value target Critic-target network Q′ obtains the target Q value Q′(s i ,μ′(s i |θ μ′ )|θ Q′ ), s i Represents the state space obtained at the i-th iteration and regularly copies the Q value of the current Critic-online network Q network parameter θ Q To update its own network parameters θ Q′ .

[0047] The network parameters θ of the Q-value target Critic-target network Q′ Q′ and the network parameters θ of the action target Actor-target network μ′μ′ The updates are as follows:

[0048] θ Q′ ←τθ Q +(1-τ)θ Q′

[0049] θ μ′ ←τθ μ +(1-τ)θ μ′

[0050] Where τ is the soft update coefficient.

[0051] Step 2) The generated voltage, frequency, and power of the tidal energy generation system are constructed as a state space and processed by the deep deterministic policy gradient DDPG algorithm and the network parameters of the deep deterministic policy gradient DDPG algorithm are updated as follows:

[0052] Step 2.1) Use random initialization to initialize the network parameters θ of the current Actor-online network μ of the deep deterministic policy gradient DDPG algorithm μ and Q value of the current Critic-online network Q network parameters θ Q And the experience replay pool D, and initialize the network parameters θ of the current Actor-online network μ μ The network parameters θ copied to the action target Actor-target network μ′ μ′ , the network parameter θ of the current Critic-online network Q after the initialization Q value Q The network parameters θ copied to the Q-value target Critic-target network Q′ Q′ .

[0053] Step 2.2) Initialize the random noise E of the deep deterministic policy gradient DDPG algorithm and obtain the current state space s of the tidal energy generation system t , based on the initialized random noise E and the current state space s t Get the current action a t , a t =μ(s t |θ μ )+E.

[0054] Step 2.3) The tidal energy generation system executes the current action a t , get the current reward r t and the next state space s t+1 , the current state space s of the tidal energy generation system t 、Current action a t 、Current reward r tand the next state space together to form experience data (s t , t , t , t+1 ) and store them in the experience replay pool D, update the current state space s t to the next state space s t+1 .

[0055] Step 2.4) randomly sample M experience data from the experience replay pool D, and calculate the target value y, calculate the Q value of the Q value current Critic-online network Q according to the target value y, and update the network parameters θ Q of the Q value current Critic-online network Q using the minimum loss function L, and perform gradient backpropagation according to the updated network parameters θ Q of the Q value current Critic-online network Q to update the network parameters θ μ of the current Actor-online network μ.

[0056] Step 2.5) the action target Actor-target network μ' copies the network parameters θ μ of the action current Actor-online network μ to update its own network parameters θ μ′ ; the Q value target Critic-target network Q' copies the network parameters θ Q of the Q value current Critic-online network Q to update its own network parameters θ Q′ .

[0057] Step 2.6) after updating, the action of the tidal current power generation system is obtained, including the amplitude, frequency and phase of the excitation current of the inverter controller (5) of the tidal current power generation system.

[0058] The method of the present application utilizes deep reinforcement learning technology, realizes real-time optimization control of the power generation voltage, frequency and power of the tidal current power generation system through online learning of the control strategy, and uses the Actor-Critic structure of the deep deterministic policy gradient DDPG algorithm. The agent learns by observing the system state, the action taken and the reward function designed. The Actor network generates control actions to optimize the power generation voltage, frequency and power of the tidal current power generation system, and the Critic network evaluates the Actor network according to the Q value of the system. The experience replay mechanism is introduced, the experience data samples generated by the interaction between the Actor network and the environment are stored in the experience replay pool, a batch of data samples are extracted for online learning to update the network parameters, and thus the optimal control strategy is generated.

[0059] Step 3) Determine whether the state space of the current tidal energy power generation system has reached the terminal state. If not, continue with the next round of learning, input the state space of the tidal energy power generation system into the deep deterministic policy gradient DDPG algorithm after the network parameters are updated, and repeat steps 2)-3) until the state space of the tidal energy power generation system reaches the terminal state. Save the network parameters and weights of the current deep deterministic policy gradient DDPG algorithm, and control the tidal energy power generation system in real time through the actions obtained by the current deep deterministic policy gradient DDPG algorithm. Then, in real time, optimize the learning parameters to perform intelligent power generation quality control on the actions of the tidal energy power generation system. Specifically, when the optimized learning parameter P exceeds the preset parameter threshold P threshold When , re-evaluation and training are performed, the current state space of the tidal energy generation system is input into the deep deterministic policy gradient DDPG algorithm after the network parameters are updated and steps 2)-3) are repeated until the optimized learning parameter P does not exceed the preset parameter threshold P threshold , thereby intelligently controlling the power generation quality of the tidal energy power generation system through current actions.

[0060] like Figure 2 As shown, in a specific implementation, the method of the present invention is used to perform real-time control of a doubly-fed induction generator system. The doubly-fed induction generator system includes a gearbox 1, a doubly-fed induction generator 2, a bidirectional frequency converter 3, a power grid 4, an inverter controller 5, and an operating status detector 6. The stator winding of the doubly-fed induction generator 2 is excited by three symmetrical power sources with fixed frequencies, while the rotor winding is excited by a three-phase power source with adjustable frequencies. The inverter controller 5, through the control of the low-power components of the inverter circuit by the bidirectional frequency converter 3, can change the amplitude, frequency, and phase angle of the rotor excitation current of the doubly-fed induction generator 2. By changing the excitation frequency, the speed can be adjusted. This allows the doubly-fed induction generator 2 to rapidly change its speed when the load suddenly changes, fully utilizing the rotor's kinetic energy and releasing and absorbing the load. This significantly reduces the disturbance to the power grid 4 compared to conventional motors. Furthermore, by adjusting the amplitude and phase of the rotor excitation current to regulate active and reactive power, the unit's efficiency is improved while also stabilizing the frequency and voltage of the power grid 4. The parameters of the state space s = [v, f, p] of the doubly fed induction power generation system are obtained by the operation state detector 6, which detects various data and parameters of the doubly fed induction power generation system in real time. The action space a = [I m ,I f ,I p ], where I m Indicates the magnitude of the excitation current of the inverter controller 5, I f is the frequency of the excitation current, I pFor the phase of the excitation current, the voltage, frequency and active power of the double-fed induction power generation system can be controlled by the excitation current, the double-fed induction generator 2 is applied to tidal current power generation, and problems such as unadjustable motor rotating speed and low unit efficiency can be solved. Meanwhile, since the double-fed motor is adjustable for reactive power and active power, the double-fed motor can play a role in voltage stabilization and frequency stabilization for the power grid, and the power generation quality is improved. When the double-fed induction power generation system is controlled, the final state space s T = [v target ,f target ,p target ] is set, and the state space can be set as a floating interval.

[0061] The method has certain universality, and each parameter in the corresponding state space and action space can be modified according to specific needs, for example, the flow rate and rotating speed are added, and the parameters are appropriately modified according to actual needs and effects.

[0062] Although the embodiments of the present application have been shown and described, it can be understood by those skilled in the art that various changes, modifications, replacements and modifications can be made to the embodiments without departing from the principles and spirits of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A method for intelligently controlling power generation quality of a tidal energy power generation system, characterized in that: include: Step 1) Build a deep deterministic policy gradient (DDPG) algorithm based on optimized learning parameters; Step 2) The generated voltage, frequency, and power of the tidal energy generation system are constructed into a state space and processed by the deep deterministic policy gradient DDPG algorithm and the network parameters of the deep deterministic policy gradient DDPG algorithm are updated to obtain the action of the tidal energy generation system after processing; Step 3) Determine whether the current state space of the tidal energy power generation system has reached a terminal state. If not, input the state space of the tidal energy power generation system into the deep deterministic policy gradient DDPG algorithm after the network parameters are updated and repeat steps 2)-3) until the state space of the tidal energy power generation system reaches a terminal state. Use the actions obtained by the current deep deterministic policy gradient DDPG algorithm to control the tidal energy power generation system in real time. Then, optimize the learning parameters in real time to intelligently control the power generation quality of the tidal energy power generation system. In the step 1), optimize the learning parameters The details are as follows: in, and They are the generation voltages of the tidal energy generation system The mean absolute error and standard deviation The weight coefficient of and The power generation frequency of the tidal energy generation system is The mean absolute error and standard deviation The weight coefficient of and are the power generation power of the tidal energy generation system The mean absolute error and standard deviation The weight coefficient of and are the rewards at time tn of the deep deterministic policy gradient DDPG algorithm within n time steps. and the reward at the current time t The reward change between Reward changes The weight coefficient of The reward at the current time t of the deep deterministic policy gradient DDPG algorithm The details are as follows: in, and are the power generation voltage and target value of the tidal energy generation system at the current time t, and are the power generation frequency and target value of the tidal energy power generation system at the current time t, and are the power generation power and target value of the tidal energy generation system at the current time t, 、 and They are the generation voltages of the tidal energy generation system ,frequency and power The weight coefficient of In the step 3), the power generation quality of the tidal energy power generation system is intelligently controlled by optimizing the learning parameters in real time. Specifically, when optimizing the learning parameters Exceeding the preset parameter threshold When the current state space of the tidal energy generation system is input into the deep deterministic policy gradient DDPG algorithm after the network parameters are updated, and steps 2)-3) are repeated until the learning parameters are optimized. Does not exceed the preset parameter threshold , thereby intelligently controlling the power generation quality of the tidal energy power generation system through current actions.

2. The method for intelligently controlling power generation quality of a tidal energy power generation system according to claim 1, characterized in that: In step 1), the deep deterministic policy gradient DDPG algorithm includes the action current Actor-online network , Actor-target network , Q value current critic-online network , Q-value target Critic-target network and experience replay pool, action current Actor-online network Perform your own network parameters Iterative update, and according to the state space of the tidal energy generation system Generate current action , to interact with the environment of the tidal energy generation system to generate the next state space and current rewards ; Actor-target network The next spatial state sampled from the experience replay pool Generate the optimal next action , and periodically replicate actions to the current Actor-online network Network parameters To update its own network parameters ; Q value current critic-online network Perform your own network parameters Iterative update and obtain the current Value; Q value target Critic-target network Get the target Q value , Represents the state space obtained at the i-th iteration and periodically copies the Q value of the current Critic-online network Network parameters To update its own network parameters .

3. The method for intelligently controlling power generation quality of a tidal energy power generation system according to claim 2, characterized in that: In step 2), the generated voltage, frequency, and power of the tidal energy generation system are constructed as a state space and processed by the deep deterministic policy gradient DDPG algorithm and the network parameters of the deep deterministic policy gradient DDPG algorithm are updated as follows: Step 2.1) Initialize the current Actor-online network of the Deep Deterministic Policy Gradient (DDPG) algorithm using random initialization Network parameters and Q value of the current Critic-online network Network parameters And the experience replay pool D, and initialize the current Actor-online network Network parameters Copy to the Actor-target network Network parameters , the initialized Q value is used in the current Critic-online network Network parameters Copy to Q value target Critic-target network Network parameters ; Step 2.2) Initialize the random noise E of the deep deterministic policy gradient DDPG algorithm and obtain the current state space of the tidal energy generation system , based on the initialized random noise E and the current state space Get the current action , ; Step 2.3) The tidal energy generation system executes the current action , get the current reward and the next state space , the current state space of the tidal energy generation system , Current Action , Current Rewards and the next state space are constructed as empirical data And store it in the experience replay pool D, and convert the current state space Update to the next state space ; Step 2.4) Randomly sample M experience data from the experience replay pool D and calculate the target value y. Calculate the Q value based on the target value y. Current Critic-online network of Value, thereby using the minimized loss function L to update the Q value current Critic-online network Network parameters , according to the updated Q value of the current Critic-online network Network parameters Perform gradient backpropagation to update the current Actor-online network Network parameters ; Step 2.5) Actor-target network Copy action current Actor-online network Network parameters To update its own network parameters ; Q-value target Critic-target network Copy the Q value of the current Critic-online network Network parameters To update its own network parameters ; Step 2.6) After the update is completed, the action of the tidal energy generation system is obtained, including the amplitude, frequency and phase of the excitation current of the inverter controller (5) of the tidal energy generation system.

Images