Road domain new energy frequency stability control method considering new energy uncertainty

Abstract

The invention discloses a road area new energy frequency stability control method considering uncertainty of new energy, and relates to the technical field of road area new energy, comprising the following steps: collecting data of a road area new energy power generation system, and constructing an inertial response controller parameter optimization problem model; constructing a controller parameter adaptive optimization additional learning structure based on adaptive dynamic programming (ADP), and calculating an optimal control parameter; training and constructing an evaluation network CNN online algorithm and an execution network ANN online algorithm based on a gradient descent method; and analyzing the convergence of the inertial response controller parameter optimization method, and verifying the effectiveness of the inertial response controller parameter optimization method. According to the method, an inertia response controller parameter optimization and self-adaptive adjustment technology is provided, the frequency stability of a road domain power system is effectively improved, controller parameters are optimized on line on the premise that the rotating speed is not out of limit, and the primary frequency modulation supporting capacity of a road domain doubly-fed fan is provided.

Description

Technical Field

[0001] The present invention relates to the field of road-related new energy technology, and in particular to a method for controlling the frequency stability of road-related new energy taking into account the uncertainty of new energy. Background Art

[0002] The output volatility and operating condition uncertainty of distributed renewable energy in the road area are significant, which makes the system frequency regulation change from uncertainty only on the load side to uncertainty on both the generation side and the load side. Traditional power systems generally use the method of reserving sufficient active reserve to ensure frequency stability, but with the continuous increase in the scale of distributed renewable energy installed capacity, the system active reserve demand is also increasing, which brings new challenges to the frequency stability of the power system.

[0003] There are two ways to improve the primary frequency regulation support capacity of distributed renewable energy in the road area. One is to add energy storage systems to distributed renewable energy in the road area, and use the advantages of flexible control, stable performance, rapid and accurate regulation of energy storage systems to mitigate active impact and increase the frequency regulation capacity of wind farms. However, no matter which control method is adopted, the joint frequency regulation method inevitably requires the addition of additional energy storage equipment, and the economic issues between the energy storage backup capacity and the loss caused by frequency deviation need to be considered, and reasonable configuration is difficult. Summary of the invention

[0004] In view of the problem that when frequency control is performed on road-area renewable energy, additional energy storage equipment needs to be added in the prior art, which makes reasonable configuration difficult, and the present invention is proposed.

[0005] Therefore, the problem to be solved by the present invention is how to provide a method for studying the optimization of control parameters of a road-area renewable energy power generation system and adaptive regulation technology based on adaptive dynamic programming ADP theory.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In the first aspect, an embodiment of the present invention provides a method for frequency stability control of new energy in the road domain taking into account the uncertainty of new energy, which includes collecting data of the new energy power generation system in the road domain, constructing an inertial response controller parameter optimization problem model; constructing an additional learning structure for adaptive optimization of controller parameters based on adaptive dynamic programming ADP and calculating the optimal control parameters; constructing an evaluation network CNN online algorithm and an execution network ANN online algorithm based on gradient descent training; analyzing the convergence of the inertial response controller parameter optimization method, and verifying the effectiveness of the inertial response controller parameter optimization method.

[0008] As a preferred solution of the road domain new energy frequency stability control method considering the uncertainty of new energy described in the present invention, wherein: the controller parameter adaptive optimization additional learning structure based on adaptive dynamic programming ADP includes an evaluation network CNN online algorithm and an execution network ANN online algorithm; the evaluation network CNN online algorithm obtains the direct mapping relationship between the system state variables and the control variables and the cost function by training the neural network weights; the execution network ANN online algorithm obtains the mapping relationship between the system state variables and the approximate optimal cost function obtained by the evaluation network CNN online algorithm and the system control variable increment by training the neural network.

[0009] As a preferred solution of the road domain new energy frequency stability control method considering the uncertainty of new energy described in the present invention, wherein: the calculation of the optimal control parameters includes the following steps: before the iteration has not converged, the output of the evaluation network CNN online algorithm is an approximate cost function The execution network ANN online algorithm obtains the output variable of the controller through the state variable, and approximates the optimal control variable under the optimal cost through the neural network. The specific formula is as follows:

[0010] K(t+1)=K(t)+ΔK(t)=K(t)+Mu(t)

[0011] Among them, K(t) is the control variable when time is t; M is the weighting coefficient; u(t) is the optimal control variable; the evaluation network CNN online algorithm and the execution network ANN online algorithm are repeatedly iterated to calculate the optimal control parameters.

[0012] As a preferred solution of the road domain new energy frequency stability control method considering the uncertainty of new energy described in the present invention, the specific mapping relationship between the input and output layers of the evaluation network CNN online algorithm is as follows:

[0013]

[0014]

[0015]

[0016] Among them, q is the input of the middle hidden layer neuron; p is the output of the middle hidden layer neuron; N hc is the number of hidden layer neurons; is the weight between the input layer and the hidden layer; is the weight between the input and output neurons of the hidden layer; is a common activation function in neural networks; u j-n(t) is the jnth element in the output of the execution network ANN online algorithm; the specific mapping relationship between the input and output layers of the execution network ANN online algorithm is as follows:

[0017]

[0018]

[0019]

[0020] Among them, g is the input of the middle hidden layer neuron; h is the output of the hidden layer neuron; N ha is the number of hidden layer neurons; is the mapping relationship between input neurons and hidden layer neurons; is the mapping relationship between the hidden layer neurons and the kth output neuron.

[0021] As a preferred solution of the road domain new energy frequency stability control method considering the uncertainty of new energy according to the present invention, the training results of the evaluation network CNN online algorithm are as follows:

[0022]

[0023] in, is the approximate cost function; To evaluate the network weights from the hidden layer to the output layer of the CNN online algorithm An estimated value of The vector form of the hidden layer neuron output of the evaluation network CNN online algorithm; the training results of the execution network ANN online algorithm are as follows:

[0024]

[0025] Among them, u(t) is the optimal control variable; The network weights from the hidden layer to the output layer for executing the ANN online algorithm An estimated value of It is the vector form of the hidden layer neuron output of the ANN online algorithm. The calculation formula of the optimal parameter obtained after repeated iterations is as follows:

[0026]

[0027] in, is the optimal parameter obtained after repeated iterations; the training deviation e of the evaluation network CNN online algorithm c The specific formula of (t) is as follows:

[0028]

[0029] The training objective E of the evaluation network CNN online algorithm is c The specific formula of (t) is as follows:

[0030]

[0031] The specific formula of the CNN online algorithm based on the gradient descent method for training and evaluating the network is as follows:

[0032]

[0033] Among them, l c is the weight coefficient in the process of training CNN; the training deviation e of the execution network ANN online algorithm a The specific formula of (t) is as follows:

[0034]

[0035] The training objective E of the execution network ANN online algorithm is a The specific formula of (t) is as follows:

[0036]

[0037] The specific formula of the ANN online algorithm based on gradient descent training execution network is as follows:

[0038]

[0039] Among them, l a is the weight coefficient in the ANN training process; C(t) is a matrix; the calculation formula of the elements in the matrix is ​​as follows:

[0040]

[0041] Among them, C ij (t) is the element in the i-th row and j-th column of the matrix; φ ci (t) is φ c The i-th element in (t); To evaluate the network weights connecting the i-th input node and the n+j-th hidden layer node in the CNN online algorithm.

[0042] As a preferred solution of the frequency stability control method of new energy in the road domain considering the uncertainty of new energy described in the present invention, the instantaneous cost function r(t) includes a positive definite function r(t) composed of the square of the frequency deviation. f (t), the positive definite function r consisting of the quadratic form related to the control variable u (t) and the positive definite function r constraining the wind rotor speed w (t); The calculation formula of the instantaneous cost function r(t) is as follows:

[0043] r(t)=(r f (t)+r w (t)+r u (t))Δt

[0044] The square of the frequency deviation constitutes a positive definite function r f The specific formula of (t) is as follows:

[0045] r f (t) = Q f (ff ref ) 2

[0046] The quadratic form associated with the control variable constitutes a positive definite function r u The specific formula of (t) is as follows:

[0047] r u (t) = u T R u u

[0048] The positive definite function r constraining the wind rotor speed w The specific formula of (t) is as follows:

[0049]

[0050] Among them, Q f >0 is a positive constant; is a symmetric positive definite matrix; u is the control variable; M1 is the transition cost; M2 is the cost function when the speed exceeds the safe operating range.

[0051] As a preferred solution of the road domain new energy frequency stability control method considering the uncertainty of new energy described in the present invention, preprocessing is required before the system frequency and the rotor speed of the doubly fed wind turbine are input into the controller parameter adaptive optimization structure based on ADP; the specific formula of the preprocessing is as follows:

[0052] x1=ff ref

[0053]

[0054] Among them, f is the system frequency; w is the rotor speed of the doubly fed wind turbine; x1 is the preprocessed parameter of the system frequency; x2 is the preprocessed parameter of the rotor speed of the doubly fed wind turbine.

[0055] On the second aspect, in order to further solve the safety problems existing in the frequency stability control of new energy in the road domain, the embodiment of the present invention provides a road domain new energy frequency stability control system taking into account the uncertainty of new energy, which includes: a model construction module, which is used to collect road domain new energy power generation system data and construct relevant models of inertial response controller parameter optimization problems; a parameter calculation module, which is used to construct an additional learning structure for adaptive optimization of controller parameters based on adaptive dynamic programming ADP, and calculate the optimal control parameters through repeated iterations; an algorithm training module, which is used to train the evaluation network CNN online algorithm and the execution network ANN online algorithm; a practical application module, which is used to design the cost function in the actual environment according to the constructed additional learning structure for adaptive optimization of controller parameters based on adaptive dynamic programming ADP, and pre-process the parameters.

[0056] In a third aspect, an embodiment of the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, it implements any step of a method for controlling the frequency stability of new energy in a road domain taking into account the uncertainty of new energy as described in the first aspect of the present invention.

[0057] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of a method for controlling the frequency stabilization of new energy on a road surface taking into account the uncertainty of new energy as described in the first aspect of the present invention.

[0058] The beneficial effects of the present invention are as follows: the present invention designs a primary frequency regulation control strategy for a doubly-fed wind turbine in a road domain that participates in the primary frequency regulation of the system through an inertial response controller based on the direct heuristic approximate dynamic programming theory; proposes an inertial response controller parameter optimization and adaptive adjustment technology, effectively improves the frequency stability of the road domain power system, optimizes the controller parameters online while ensuring that the speed does not exceed the limit, and provides a primary frequency regulation support capability for the doubly-fed wind turbine in the road domain. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. Among them:

[0060] Figure 1 This is a structural diagram of the controller parameter adaptive optimization based on ADP in Example 1.

[0061] Figure 2 This is a graph of the Sigmod excitation function in Example 1.

[0062] Figure 3 This is the evaluation network structure diagram in Example 1.

[0063] Figure 4 This is a diagram of the network structure executed in Example 1.

[0064] Figure 5 is the positive definite function r of the wind wheel speed w in Example 1 w The relationship curve diagram.

[0065] Figure 6 This is a structural diagram of the test system in Example 2.

[0066] Figure 7 is K in Example 2 p Plot of the changes during 35 runs.

[0067] Figure 8 is K in Example 2 d Plot of the changes during 35 runs.

[0068] Fig. 9 This is a cost function change curve diagram of Example 1 in Example 2.

[0069] Fig.10 This is a frequency response curve of 10% load before and after parameter optimization of Example 1 in Example 2.

[0070] Fig.11 This is a rotor speed response curve diagram before and after parameter optimization of Example 1 in Example 2.

[0071] Fig.12 This is a frequency response curve diagram during the parameter optimization process of Example 1 in Example 2.

[0072] Fig.13 K is the second example in Example 2 p Plot of the changes during 35 runs.

[0073] Fig.14 K is the second example in Example 2 d Plot of the changes during 35 runs.

[0074] Fig.15 This is a curve diagram of the cost function change of Example 2 in Example 2.

[0075] Fig.16 This is a frequency response curve of 10% load before and after parameter optimization of Example 2 in Example 2.

[0076] Fig.17 This is a rotor speed response curve diagram before and after parameter optimization of Example 2 in Example 2. DETAILED DESCRIPTION

[0077] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0078] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0079] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0080] Example 1

[0081] Reference Figure 1 to Figure 5 , which is the first embodiment of the present invention, and provides a method for controlling the frequency stability of new energy in a road domain considering the uncertainty of new energy, including the following steps:

[0082] S1: Collect data on the road-related new energy power generation system and build a parameter optimization model for the inertial response controller.

[0083] Preferably, data of the road-related new energy power generation system is collected, and a model is built for the inertial response controller parameter optimization problem.

[0084] Preferably, modeling for the inertial response controller parameter optimization problem includes a discrete-time nonlinear system and a parameterized controller.

[0085] Specifically, the specific formula for discrete-time nonlinear systems is as follows:

[0086] x(t+1)=F(x(t),v(t))

[0087] The specific formula of the parameterized controller is as follows:

[0088] v(t)=h(x(t),K)

[0089] Where t is the discrete time; x is the state vector, v is the control vector, K is the controller parameter vector, F(x(t),v(t)) is the system dynamics, h(x(t),K) is the control strategy,

[0090] Furthermore, the goal of inertial response controller parameter optimization is to optimize controller performance without losing system stability, that is, to maintain the desired state variables or control variables within a certain range.

[0091] Specifically, the calculation formula of the instantaneous cost function r(t) at time t is as follows:

[0092] r(t)=W(x(t))+v(t) T Rv(t)

[0093] Among them, W(x(t)) is the state variable cost, which represents the restriction on the state variable. R is the weight matrix, which represents the restriction on the control vector. Generally, we do not want the control variable frequency to jump.

[0094] Furthermore, the instantaneous cost function r(t) is used to evaluate the deviation between the current system operating state and the expected operating state. At the same time, the instantaneous cost function r(t) needs to be frequently calculated in both policy iteration and cost function iteration.

[0095] Preferably, the calculation formula of the cost function is as follows:

[0096]

[0097] Among them, α is the time-sharing weight, 0<α<1; J(t) is the sum of r(t) of all operating states of the entire system between the current moment and the infinite moment. Since the future system dynamics cannot be accurately obtained, J(t) is generally solved through iteration to obtain an approximate value and cannot be calculated directly.

[0098] Furthermore, optimizing the controller parameters is to find the optimal controller with the optimal parameter K as the control parameter, and to find the minimum value of the cost function while keeping the system from losing stability.

[0099] Specifically, the specific formula of the cost function related to the controller parameter vector K is as follows:

[0100]

[0101] Further, we get J * The specific formula of the controller parameter vector after (t) is as follows:

[0102]

[0103] Among them, J * (t) is the cost function associated with the controller parameter vector K; K * To get J * (t) is the controller parameter vector after .

[0104] S2: Construct an additional learning structure for adaptive optimization of controller parameters based on adaptive dynamic programming (ADP) and calculate the optimal control parameters.

[0105] Preferably, the controller parameter adaptive optimization additional learning structure based on adaptive dynamic programming ADP includes an evaluation network CNN online algorithm and an execution network ANN online algorithm.

[0106] Furthermore, the specific approximation structures of the evaluation network CNN online algorithm and the execution network ANN online algorithm are neural networks.

[0107] Specifically, Figure 1 The figure shows the adaptive optimization structure of controller parameters based on ADP. The solid line represents the data transmission path adopted and optimized in the iteration, and the dotted line represents the path for data exchange between the two neural networks.

[0108] Specifically, the evaluation network CNN online algorithm obtains the direct mapping relationship between the system state variables and control variables and the cost function by training the neural network weights.

[0109] The execution network ANN online algorithm obtains the mapping relationship between the system state variables by training the neural network and the approximate optimal cost function obtained by the evaluation network CNN online algorithm and the system control variable increment.

[0110] Furthermore, the evaluation network CNN online algorithm uses a neural network to simulate the evaluation function, fits the optimal cost function that originally requires reverse time order recursion through state variables and control variables, and then calculates the cost function J(t) that meets the definition, thereby improving the efficiency of the algorithm.

[0111] Specifically, the calculation of the optimal control parameters includes the following steps: before the iteration converges, the output of the CNN online algorithm is evaluated as an approximate cost function

[0112] The training objective of the evaluation network CNN online algorithm satisfies the Bellman equation. The specific formula is as follows:

[0113] J(t-1)=r(t-1)+αJ(t)

[0114] Where t is discrete time; r(t-1) is the instantaneous cost function at time t-1; α is the time-sharing weight; J(t) is the sum of the instantaneous cost functions r(t) of all operating states of the entire system between the current moment and the infinite moment.

[0115] The execution network ANN online algorithm uses the state variables to obtain the output variables of the controller, and approximates the optimal control variables under the optimal cost through the neural network. The specific formula is as follows:

[0116] K(t+1)=K(t)+ΔK(t)=K(t)+Mu(t)

[0117] Among them, K(t) is the control variable at time t; M is the weighting coefficient, u(t) is the optimal control variable; in order to eliminate static errors, the proportional-integral structure is used to complete iterative feedback in the parameter update link; the iterative goal of the execution network ANN online algorithm is to change different system control variables to make the evaluation network CNN online algorithm train the approximate cost function As small as possible, and because the cost functions are all positive definite functions, the training target of the execution network ANN online algorithm is generally set to U C (t)=0.

[0118] The evaluation network CNN online algorithm and the execution network ANN online algorithm are iterated repeatedly. When the iteration converges, the optimal control parameters under the required performance indicators are obtained.

[0119] Furthermore, since the evaluation network CNN online algorithm uses state variables and output variables as data sets during training, the approximate cost function obtained by training It is directly related to the state variables and output variables, so the controller parameter adaptive optimization based on adaptive dynamic programming (ADP) with additional learning structure can optimize the parameters when the system model is unknown.

[0120] S3: Based on the gradient descent method, the evaluation network CNN online algorithm and the execution network ANN online algorithm are constructed.

[0121] Preferably, the evaluation network CNN online algorithm and the execution network ANN online algorithm used in the present invention are both composed of neural networks with similar structures and only one hidden layer, and the hidden layer nodes of the neural network use Sigmund excitation function.

[0122] Specifically, Figure 2 The figure shows a curve graph of the Sigmod activation function, from which we can see that the Sigmod activation function changes slowly, and can reflect slight changes in the input in slight changes in the output. It is a common activation function in neural networks.

[0123] Furthermore, the specific formula of the Sigmod activation function is as follows:

[0124]

[0125] in, is the Sigmod activation function.

[0126] Preferably, Figure 3The figure shows the network structure of the evaluation network CNN online algorithm. The specific mapping relationship between the input and output layers of the evaluation network CNN online algorithm is as follows:

[0127]

[0128]

[0129]

[0130] Among them, q is the input of the middle hidden layer neuron; p is the output of the middle hidden layer neuron; N hc is the number of hidden layer neurons; is the weight between the input layer and the hidden layer, indicating the influence of the j-th input neuron on the i-th hidden layer neuron; is the weight between the input and output neurons of the hidden layer; u j-n (t) is the jnth element in the output of the ANN online algorithm.

[0131] Preferably, Figure 4 The network structure of the execution network ANN online algorithm is shown. The specific mapping relationship between the input and output layers of the execution network ANN online algorithm is as follows:

[0132]

[0133]

[0134]

[0135] Among them, g is the input of the middle hidden layer neuron; h is the output of the hidden layer neuron; N ha is the number of hidden layer neurons; is the mapping relationship between input neurons and hidden layer neurons; is the mapping relationship between the hidden layer neurons and the kth output neuron.

[0136] Furthermore, at the beginning of each training, the weights between the input layer neurons and the hidden layer neurons of the evaluation network CNN online algorithm and the execution network ANN online algorithm are set to 1, and the weights between the hidden layer neurons and the output layer neurons are randomly assigned between (0,1), and the outputs of the hidden layer neurons of the evaluation network CNN online algorithm and the execution network ANN online algorithm are rewritten into vector form. and

[0137] Specifically, the training results of the evaluation network CNN online algorithm are as follows:

[0138]

[0139] in, is the approximate cost function; To evaluate the network weights from the hidden layer to the output layer of the CNN online algorithm An estimated value of It is the vector form of the hidden layer neuron output of the evaluation network CNN online algorithm.

[0140] Specifically, the training results of executing the network ANN online algorithm are as follows:

[0141]

[0142] Among them, u(t) is the optimal control variable; The network weights from the hidden layer to the output layer for executing the ANN online algorithm An estimated value of It is the vector form of the output of the hidden layer neurons that execute the ANN online algorithm.

[0143] Specifically, the calculation formula of the optimal parameter obtained after repeated iterations is as follows:

[0144]

[0145] in, is the optimal parameter obtained after repeated iterations.

[0146] Furthermore, the training bias of the CNN online algorithm will be evaluated. c (t) is set to the form of Bellman residual, the specific formula is as follows:

[0147]

[0148] Furthermore, the training objective E of the online CNN algorithm is evaluated. c (t) is set to the quadratic form of the Bellman residual amplitude, and the specific formula is as follows:

[0149]

[0150] Furthermore, the specific formula of the online algorithm for training the evaluation network CNN based on the gradient descent method is as follows:

[0151]

[0152] Among them, l c is the weight coefficient in the process of training CNN.

[0153] Furthermore, the training deviation of the ANN online algorithm is executed a The specific formula of (t) is as follows:

[0154]

[0155] Furthermore, the training objective E of the execution network ANN online algorithm is a The specific formula of (t) is as follows:

[0156]

[0157] Furthermore, the specific formula of the ANN online algorithm for training the execution network based on the gradient descent method is as follows:

[0158]

[0159] Among them, l a is the weight coefficient in the process of training ANN; C(t) is a matrix.

[0160] Specifically, the calculation formula for the elements in the matrix is ​​as follows:

[0161]

[0162] Among them, C ij (t) is the element in the i-th row and j-th column of the matrix, φ ci (t) is φ c The i-th element in (t); To evaluate the network weights connecting the i-th input node and the n+j-th hidden layer node in the CNN online algorithm.

[0163] S4: Analyze the convergence of the inertial response controller parameter optimization method and verify the effectiveness of the inertial response controller parameter optimization method.

[0164] Preferably, the optimization object of the parameter optimization method proposed above is selected as the PD inertia controller parameter K p and K d The training data sets for the evaluation network CNN online algorithm and the execution network ANN online algorithm are the grid connection point frequency f and the wind rotor speed w of the doubly fed wind turbine.

[0165] Furthermore, the instantaneous cost function r(t) includes the positive definite function r(t) consisting of the square of the frequency deviation f (t), the positive definite function r consisting of the quadratic form related to the control variable u (t) and the positive definite function r constraining the wind rotor speed w (t).

[0166] Specifically, the calculation formula of the instantaneous cost function r(t) is as follows:

[0167] r(t)=(r f (t)+r w (t)+ru (t))Δt

[0168] Specifically, considering the frequency stability of the grid-connected point system, the square of the frequency deviation is used to form a positive definite function r f (t), the specific formula is as follows:

[0169] r f (t) = Q f (ff ref ) 2

[0170] Specifically, in order to prevent the system from becoming unstable due to the frequency jump of the system control variable, the quadratic form related to the control variable is used to form a positive definite function r u (t), the specific formula is as follows:

[0171] r u (t) = u T R u u

[0172] Specifically, Figure 5 The figure shows the relationship between the wind wheel speed w and the positive definite function r w In order to prevent the rotor speed from dropping too much when participating in frequency modulation and triggering the speed protection, the positive definite function r is used. w (t) Constrain the wind rotor speed w, the specific formula is as follows:

[0173]

[0174] Among them, Q f >0 is a positive constant; is a symmetric positive definite matrix; u is the control variable; M1 is the transition cost; M2 is the cost function when the speed exceeds the safe operating range. When the speed exceeds the safe operating range, the cost function M2>0.

[0175] Preferably, preprocessing is required before the system frequency f and the doubly-fed wind turbine rotor speed w are input into the ADP-based controller parameter adaptive optimization structure.

[0176] Specifically, the specific formula for preprocessing is as follows:

[0177] x1=ff ref

[0178]

[0179] Among them, f is the system frequency; w is the rotor speed of the doubly fed wind turbine; x1 is the preprocessed parameter of the system frequency; x2 is the preprocessed parameter of the rotor speed of the doubly fed wind turbine.

[0180] This embodiment also provides a road-domain new energy frequency stability control system that takes into account the uncertainty of new energy, including: a model construction module, used to collect road-domain new energy power generation system data and construct relevant models for the inertial response controller parameter optimization problem; a parameter calculation module, used to construct an additional learning structure for adaptive optimization of controller parameters based on adaptive dynamic programming ADP, and calculate the optimal control parameters through repeated iterations; an algorithm training module, used to train the evaluation network CNN online algorithm and the execution network ANN online algorithm; a practical application module, used to design the cost function in the actual environment according to the constructed additional learning structure for adaptive optimization of controller parameters based on adaptive dynamic programming ADP, and pre-process the parameters.

[0181] This embodiment also provides a computer device, which is suitable for a method for stabilizing the frequency of new energy on a road surface taking into account the uncertainty of new energy, and includes: a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement a method for stabilizing the frequency of new energy on a road surface taking into account the uncertainty of new energy as proposed in the above embodiment.

[0182] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0183] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, a method for controlling the frequency stability of new energy in a road domain taking into account the uncertainty of new energy is implemented as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as a static random access memory (SRAM), an electrically erasable programmable read-only memory (EEPROM), an erasable programmable read-only memory (EPROM), a programmable read-only memory (PROM), a read-only memory (ROM), a magnetic memory, a flash memory, a disk or an optical disk.

[0184] In summary, the present invention designs a primary frequency regulation control strategy for a doubly fed wind turbine in the road domain that participates in the primary frequency regulation of the system through an inertial response controller based on the direct heuristic approximate dynamic programming theory; proposes an inertial response controller parameter optimization and adaptive adjustment technology to effectively improve the frequency stability of the road domain power system, optimizes the controller parameters online while ensuring that the speed does not exceed the limit, and provides the primary frequency regulation support capability for the doubly fed wind turbine in the road domain.

[0185] Example 2

[0186] Reference Figure 6 to Figure 17 , which is the second embodiment of the present invention. This embodiment is different from the first embodiment in that, in order to verify its beneficial effects, operating data and instructions of the present invention in an actual environment are provided.

[0187] This example takes the PD-type inertial response controller of a doubly-fed wind turbine as the research object, optimizes its control parameters through an additional learning structure based on the ADP algorithm, optimizes the output of the doubly-fed wind turbine in frequency modulation, and thus makes the frequency response of the system better.

[0188] like Figure 6 The figure shows a road-domain distributed renewable energy grid-connected power system built in MATLAB / Simulink. The system includes three synchronous generators and one wind turbine. R =0.3, K g =20, K H =7, T H =0.3, T d =0.6, H area =13,Darea =0.8, H wind =5.04, the capacity of synchronous machines is 150MW, the capacity of doubly-fed wind turbines is 150MW, and the wind speed remains unchanged at 10m / s.

[0189] During the parameter optimization process, the number of hidden neuron nodes in the evaluation network CNN online algorithm and the execution network ANN online algorithm are both 10.

[0190] In the optimization process, α=0.95, l c =0.002, l a =0.002.

[0191] In the cost function setting Q f =3, R u =diag(0.2,0.2),w ll =0.8, w cl =0.9, w cu =1.2, w lu =1.3, M1=1, M2=3.

[0192] When performing the first example, first set the PD inertia controller parameter K p With K d are all set to 0, and a random load within 0.2pu is set at bus8, so that the system suffers a step load disturbance every 60s. At the same time, the proposed parameter optimization structure continuously samples the system operation status, trains the evaluation network CNN online algorithm and the execution network ANN online algorithm, and then calculates the optimal control parameters. When the parameters converge, a step load of 0.1pu is input into bus8 to compare the system frequency responses before and after optimization to verify the effectiveness of the parameter optimization method.

[0193] In the second example, based on the optimal control parameters obtained in the first example, the synchronous generator at bus2 is cut off to change the system power flow, and the PD controller parameters are re-optimized to verify the adaptive ability of the proposed method.

[0194] like Figure 7 and Figure 8 The figure shows the numerical changes of the inertial controller parameters in the iteration process in Example 1. It can be seen that after the 32nd iteration, K p Converges around 2.6, K d Converges around 8.5.

[0195] like Fig. 9 The figure shows the change curve of the cost function in Example 1. As the number of iterations increases, the cost function continues to decrease and tends to be stable after 30 iterations.

[0196] In order to verify the effectiveness of the online parameter optimization algorithm, 20MW system active load was put into the system at t = 25s to test the frequency response of the system. Fig.10 The figure shows the frequency response curves before and after parameter optimization. It can be seen that through parameter optimization, the system frequency fluctuation is effectively suppressed, the frequency fluctuation amplitude is significantly reduced, the system frequency returns to normal 15s after the fault occurs, and the lowest frequency during the fluctuation process is improved to a certain extent.

[0197] like Fig.11 The figure shows the speed change of the doubly fed wind turbine before and after parameter optimization. The rotor speed decreases to varying degrees during the response of the inertial controller and returns to a stable operating state after the response ends. At the same time, the lowest point of the rotor speed is greater than the threshold of 0.9, which verifies the effectiveness of the algorithm.

[0198] like Fig.12 The figure shows the change of system frequency response during parameter optimization. It can be seen from the figure that the parameters that did not converge during the iteration have better frequency stability control effect than the original parameters.

[0199] It can be seen from the above example 1 that the present invention can improve the control performance of the doubly-fed wind turbine PD controller, enhance the frequency response characteristics of the system while ensuring that the speed operates within a safe range, and improve the stability and anti-interference ability of the power system frequency.

[0200] like Fig.13 and Fig.14 This is the iterative optimization process of the inertial controller parameters in Example 2. It can be seen that K p With K d All converged after 25 runs.

[0201] like Fig.15 The figure shows the change curve of the cost function in Example 2. It can be seen from the figure that as the number of iterations increases, the cost function continues to decrease, and the decrease is greater than that of the cost function in Example 1. It tends to be stable when the number of iterations is less.

[0202] like Fig.16 As shown in the figure, when the system is subjected to a 10% load disturbance, by comparing the system frequency responses when there is no inertia control and when the inertia controller uses the parameter optimization results of Example 1 with the system frequency responses when the inertia controller uses the parameter optimization results of Example 2, it is shown that the frequency of the system using the optimized parameters of Example 1 fluctuates greatly during the recovery process.

[0203] When the system power flow changes, the system operation state and working point also change accordingly, and the proportion of the output active power of the double-fed wind turbine to the active power of the power generation side increases, making K obtained in Example 1 p With K d It is no longer the optimal control parameter of the system PD controller.

[0204] like Fig.17 As shown in the figure, after using the parameters optimized again in Example 2, K p With K d The values ​​of have been reduced to a certain extent, the fan speed has been limited to the safe operating range, the phenomenon of secondary frequency drop has been eliminated, the minimum frequency of the system has been increased, and the frequency response of the system has been improved.

[0205] It can be seen from the above example 2 that after introducing the ADP-based parameter optimization structure, the doubly fed wind turbine PD controller can perform online adaptive adjustment of parameters after the stable operating point of the wind turbine changes, thereby improving the frequency response of the system and improving the stability of the system frequency while ensuring its speed operation safety index.

[0206] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for frequency stability control of new energy in road domain considering the uncertainty of new energy, characterized by: include: Collect data on new energy power generation systems in the road area and build a parameter optimization model for the inertial response controller; Construct an additional learning structure for adaptive optimization of controller parameters based on adaptive dynamic programming (ADP) and calculate the optimal control parameters; Based on the gradient descent method, the evaluation network CNN online algorithm and the execution network ANN online algorithm are constructed; The convergence of the inertial response controller parameter optimization method is analyzed, and the effectiveness of the inertial response controller parameter optimization method is verified.

2. The method for controlling the frequency stability of new energy in road domain considering the uncertainty of new energy as claimed in claim 1, characterized in that: The controller parameter adaptive optimization additional learning structure based on adaptive dynamic programming ADP includes an evaluation network CNN online algorithm and an execution network ANN online algorithm.

3. The method for controlling the frequency stability of new energy in road domain considering the uncertainty of new energy as claimed in claim 2, characterized in that: The calculation of the optimal control parameters comprises the following steps: Before the iteration converges, the output of the evaluation network CNN online algorithm is an approximate cost function The execution network ANN online algorithm obtains the output variable of the controller through the state variable, and approximates the optimal control variable under the optimal cost through the neural network. The specific formula is as follows: K(t+1)=K(t)+ΔK(t)=K(t)+Mu(t) Among them, K(t) is the control variable at time t; M is the weighting coefficient; u(t) is the optimal control variable; The evaluation network CNN online algorithm and the execution network ANN online algorithm are repeatedly iterated to calculate the optimal control parameters.

4. The method for controlling the frequency stability of new energy in road domain considering the uncertainty of new energy as claimed in claim 3, characterized in that: The specific mapping relationship between the input and output layers of the evaluation network CNN online algorithm is as follows: Among them, q is the input of the middle hidden layer neuron; p is the output of the middle hidden layer neuron; N hc is the number of hidden layer neurons; is the weight between the input layer and the hidden layer; is the weight between the input and output neurons of the hidden layer; is a common activation function in neural networks; u j-n (t) is the jnth element in the output of the online algorithm of the executing network ANN; The specific mapping relationship between the input and output layers of the execution network ANN online algorithm is as follows: Among them, g is the input of the middle hidden layer neuron; h is the output of the hidden layer neuron; N ha is the number of hidden layer neurons; is the mapping relationship between input neurons and hidden layer neurons; is the mapping relationship between the hidden layer neurons and the kth output neuron.

5. The method for controlling the frequency stability of new energy in road domain considering the uncertainty of new energy as claimed in claim 4, characterized in that: The training results of the evaluation network CNN online algorithm are as follows: in, is the approximate cost function; To evaluate the hidden layer to output layer network weights of the CNN online algorithm An estimated value of The vector form of the hidden layer neuron output of the CNN online algorithm for evaluating the network; The training results of the execution network ANN online algorithm are as follows: Among them, u(t) is the optimal control variable; The network weights from the hidden layer to the output layer for executing the ANN online algorithm An estimated value of It is the vector form of the output of the hidden layer neurons that execute the ANN online algorithm of the network; The calculation formula of the optimal parameter obtained after repeated iterations is as follows: in, is the optimal parameter obtained after repeated iterations; The training bias of the evaluation network CNN online algorithm is c (t), training target E c (t) and the specific formula of the CNN online algorithm based on gradient descent training evaluation network are as follows: Among them, l c is the weight coefficient in the process of training CNN; The training deviation of the execution network ANN online algorithm is a (t), training target E a (t), the specific formula of the ANN online algorithm for training the execution network based on the gradient descent method and the elements in the matrix are as follows: Among them, l a is the weight coefficient during training ANN; C(t) is a matrix; C ij (t) is the element in the i-th row and j-th column of the matrix; φ ci (t) is φ c The i-th element in (t); To evaluate the network weights of the network connecting the i-th input node and the n+j-th hidden layer node in the online algorithm of the network CNN.

6. The road-domain new energy frequency stability control method considering new energy uncertainty as claimed in claim 5, characterized in that: The instantaneous cost function r(t) consists of a positive definite function r consisting of the square of the frequency deviation f (t), the positive definite function r consisting of the quadratic form related to the control variable u (t) and the positive definite function r constraining the wind rotor speed w (t); The calculation formula of the instantaneous cost function r(t) is as follows: r(t)=(r f (t)+r w (t)+r u (t))Δt The square of the frequency deviation constitutes a positive definite function r f The specific formula of (t) is as follows: r f (t)=Q f (f-f ref ) 2 The quadratic form associated with the control variable constitutes a positive definite function r u The specific formula of (t) is as follows: r u (t)=u T R u u The positive definite function r constraining the wind rotor speed w The specific formula of (t) is as follows: Among them, Q f >0 is a positive constant; It is a symmetric positive definite matrix; u is the control variable; M1 is the transition cost; M2 is the cost function when the speed runs beyond the safe operating range.

7. The method for controlling the frequency stability of new energy in road domain considering the uncertainty of new energy as claimed in claim 6, characterized in that: Before the system frequency and the doubly-fed wind turbine rotor speed are input into the ADP-based controller parameter adaptive optimization structure, preprocessing is required; The specific formula of the pretreatment is as follows: x1=ff ref Among them, f is the system frequency; w is the rotor speed of the doubly fed wind turbine; x1 is the preprocessed parameter of the system frequency; x2 is the preprocessed parameter of the rotor speed of the doubly fed wind turbine.

8. A road domain new energy frequency stability control system considering the uncertainty of new energy, based on a road domain new energy frequency stability control method considering the uncertainty of new energy according to any one of claims 1 to 7, characterized in that: include, Model building module, used to collect data of road-related renewable energy power generation systems and build relevant models for inertial response controller parameter optimization problems; The parameter calculation module is used to construct an additional learning structure for controller parameter adaptive optimization based on adaptive dynamic programming (ADP), and calculate the optimal control parameters through repeated iterations; Algorithm training module, used to train the evaluation network CNN online algorithm and the execution network ANN online algorithm; The practical application module is used to design the cost function in the actual environment according to the controller parameter adaptive optimization additional learning structure based on adaptive dynamic programming ADP, and preprocess the parameters.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of a method for frequency stability control of new energy on a road surface taking into account the uncertainty of new energy are implemented as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of a method for frequency stability control of new energy on a road surface taking into account the uncertainty of new energy are implemented as described in any one of claims 1 to 7.

Images