Electric power system chaotic oscillation control method and system based on reactive power compensation device

Abstract

The invention provides an electric power system chaotic oscillation control method and system based on a reactive power compensation device. The method comprises the following steps: establishing a mathematical model of an interconnected dual-machine electric power system containing an SVC according to a kinetic equation; simplifying the mathematical model to obtain a simplified mathematical model; designing a sliding mode control function; determining a relation function between the coefficient of the exponential reaching law and the sliding hyperplane; a three-layer feedforward neural network is constructed, and training is carried out through an operation data set of the interconnected dual-computer power system containing SVC according to the relation function; the neural network calculates an adjustment value of a sliding mode control parameter and a compensation value for system uncertainty according to the input state variable; finally, the control signal acts on the interconnected dual-machine power system of the SVC, and the system state is guided to converge to a stable state. Through the combination of the neural network and sliding mode control, the resistance of the whole reactive power compensation system to the interference is further enhanced, and the stable operation and control precision of reactive power compensation are ensured.

Description

Technical Field

[0001] The present invention relates to the field of power electronics technology, and in particular to a method and system for controlling chaotic oscillation of a power system based on a reactive power compensation device. Background Art

[0002] my country is a country with a large amount of electricity consumption. With the rapid expansion of electricity consumption, my country's power grid has developed into a large or even super-large cross-regional Internet network. The concept of energy Internet provides a new idea for promoting my country's sustainable development, and provides a new method for improving energy utilization and realizing energy complementarity. Energy Internet has become the development trend of my country's power grid in the future. At the same time, the expansion of the scale of energy Internet and the continuous growth of power load have brought severe challenges to the power grid, and the stability problem of modern power system will become more prominent. Once the power system is unstable, the voltage collapse or even system decoupling caused in severe cases will cause very significant economic losses. Such incidents have occurred many times in power grids at home and abroad. When researchers and engineers initially analyzed the cause of the accident, they generally thought that the negative damping of the system caused low-frequency oscillation and then led to the instability of the power grid. However, engineers later found that in some cases, even if the damping was increased in practice, the instability phenomenon still existed.

[0003] With the deepening of research on power systems and the development of chaos theory, scholars have discovered that there is a special type of oscillation in power systems that is different from negatively damped low-frequency oscillations but has chaotic characteristics, namely chaotic oscillations. As a nonlinear dynamic behavior, chaotic oscillations have attracted the attention of researchers, and many power system instability phenomena have been found to be closely related to chaotic oscillations. Chaotic oscillations can destroy the stability of power systems, cause angle instability, frequency oscillations and voltage collapse, and even induce large-scale power outages. Therefore, in order to ensure the safe and stable operation of power systems, it is necessary to study the control methods of chaotic oscillations in power systems.

[0004] In order to suppress chaotic oscillations and avoid voltage collapse, it is a direct and effective method to compensate reactive power and damping of the power system, while the traditional reactive power compensation method is inefficient in suppressing chaotic oscillations. Static Var Compensator (SVC) based on thyristor semi-controlled power devices and Static Synchronous Compensator (STATCOM) based on IGBT, GTO and other fully controlled power devices are the best in parallel reactive power compensation devices. Their reactive power compensation effect is significant and the efficiency is extremely high, which is favored by researchers and engineers.

[0005] Variable structure sliding mode control is an important basic method of chaos control. Sliding mode variable structure control can be applied to both linear and nonlinear systems, and chaotic systems are typical nonlinear systems. Therefore, sliding mode variable structure control has been introduced into the field of chaos control and has developed rapidly. Sliding mode variable structure control can drive the moving point of the chaotic system to quickly approach the sliding surface in the approaching mode and make the chaotic system leave the non-periodic orbit and enter the expected stable state in the sliding mode. At present, sliding mode variable structure control is widely used in power system chaos control due to its excellent characteristics and significant control effect. However, the high-frequency jitter that appears near the sliding surface during the control process is still its inherent defect. How to solve this problem is the key to better realize chaotic oscillation control. Summary of the invention

[0006] The technical problem to be solved by the present invention is to provide a method and system for controlling chaotic oscillation of an electric power system based on a reactive power compensation device, so as to solve the deficiencies of the prior art described in the background technology.

[0007] To solve the above technical problems, the embodiments of the present invention provide the following technical solutions:

[0008] A method for controlling chaotic oscillation of a power system based on a reactive power compensation device comprises the following steps:

[0009] Step 1: Establish a mathematical model of an interconnected dual-machine power system containing SVC based on dynamic equations;

[0010] Step 2, simplifying the mathematical model to obtain a simplified mathematical model;

[0011] Step 3: Design a switching function based on the state variables of the simplified model, and design a variable structure sliding mode control law based on Lyapunov stability theory to obtain a sliding mode control function;

[0012] Step 4: Determine the initial state of the system and the target equilibrium point, combine the switching function and the sliding mode control function, and determine the relationship function between the coefficient of the exponential reaching law and the sliding hyperplane according to the initial state of the system and the target equilibrium point;

[0013] Step 5: construct a three-layer feedforward neural network, and train it based on the relationship function using an operating data set of an interconnected dual-machine power system containing SVC;

[0014] Step 6: The state variables of the interconnected dual-machine power system containing the SVC are fed back to the sliding mode control function and the neural network in real time. The sliding mode control function calculates a preliminary control signal according to the current state and the switching function. The neural network calculates the adjustment value of the sliding mode control parameter and the compensation value for the system uncertainty according to the input state variables.

[0015] Step 7: The sum of the preliminary control signal, the adjustment value of the sliding mode control parameter and the compensation value for the system uncertainty is used as the final control signal to act on the interconnected dual-machine power system of the SVC to guide the system state to converge to a stable state; at the same time, the state of the system is continuously monitored, and the weights of the neural network and the parameters of the sliding mode control are continuously adjusted according to new data, so that the neural network is applied to the sliding mode variable structure to approximate the relationship function between the coefficients of the exponential reaching law and the sliding hyperplane.

[0016] Optionally, the mathematical model of the interconnected dual-machine power system containing SVC is established according to the dynamic equation, specifically:

[0017] (1) Construct the rotor motion equation of the generator

[0018]

[0019] Among them, δ i , P mi , P ei 、M i are the equivalent generator electrical angle, equivalent generator mechanical power, equivalent generator electromagnetic power and inertia time constant respectively;

[0020] (2) Expanding equation (1) and establishing mathematical equations based on the second-order model of the power system, the mathematical model of the system without SVC can be obtained as follows:

[0021]

[0022] Among them, δ1 and δ2 represent the power angles of the two equivalent generators; ω1 and ω2 represent the rotor angular velocities of the two equivalent generators; ω 10 ,ω 20 represents the rated rotor angular velocity of the two equivalent generators; H1 and H2 represent the rotational inertia of the two equivalent generators; D1 and D2 represent the damping coefficients of the two equivalent generators; P m1 , P m2 Represents the mechanical power of two equivalent generators; P e1 , P e2 represents the electromagnetic power of two equivalent generators, and the electromagnetic power expression of two equivalent generators G1 and G2 is

[0023]

[0024] Where X∑ is the total reactance of the tie line between the two equivalent machines, X∑=X a +X b +X a X b B SVC , where B SVC is the susceptance of the SVC;

[0025] (3) When the model contains SVC, the closed-loop system composed of SVC and power system is generally equivalent to the first-order inertia system with closed-loop time constant. The mathematical model of SVC is:

[0026]

[0027] Among them, T, u and B SVC0 They are the inertia time constant, control quantity and initial value of susceptance of SVC respectively;

[0028] (4) Using relative power angle and relative angular velocity as state variables, let δ = δ1-δ2-δ0, ω = ω1-ω2, δ0 is the initial value of the power angle difference between the two equal generators, and substitute equation (3) into equation (2), then equation (2) can be transformed into

[0029]

[0030] (5) Let θ = -D1 / H1 = -D2 / H2, define the state variables x1 = δ, x2 = ω, and transform the system (5) into

[0031]

[0032] Also defined Process some system parameters as follows

[0033]

[0034] Taking the derivative of x3 and substituting equation (4) into it, we can get

[0035]

[0036] Optionally, the mathematical model of step 1 is simplified to obtain a simplified model, specifically:

[0037] Let f(x1,x2)=k1x2cos(x1+δ0)+k2sin(x1+δ0)

[0038] And introduce the interference term w(t), and assume that w(t) has an upper bound, that is, |w(t)|≤ρ, and get the following simplified mathematical model

[0039]

[0040] Optionally, the switching function is designed according to the state variables of the simplified model, specifically:

[0041]

[0042] Among them, s0=cx1+x2, α0 and β0 are both greater than 0, q0 and p0 are positive odd numbers, and C is a normal number.

[0043] Optionally, the variable structure sliding mode control law is designed based on Lyapunov stability theory to obtain a sliding mode control function, which is specifically:

[0044] The Lyapunov function of sliding mode control is defined as:

[0045]

[0046] Taking the first-order derivative of the Lyapunov function, we get:

[0047] in

[0048]

[0049] In order to make the sliding mode reach the condition The equivalent sliding mode control function is designed as:

[0050]

[0051] Among them, k3 is the convergence rate and C is a positive constant.

[0052] Optionally, the three-layer feedforward neural network is constructed and trained by an operation data set of an interconnected dual-machine power system containing SVC to obtain a neural network model, specifically:

[0053] The three-layer feedforward neural network has 3 neurons in the input layer, corresponding to x1, x2, and x3 of the simplified mathematical model, at least 10 neurons in the hidden layer, and 2 neurons in the output layer, respectively used to output the adjustment value of the sliding mode control parameter and the compensation value for the system uncertainty;

[0054] The simulation data of the interconnected dual-machine power system containing SVC and some actual operation data are used for training. During the training process, the weights and biases of the neural network are adjusted according to the back-propagation algorithm with adaptive learning rate until the output of the neural network meets the set accuracy requirements, that is, it can accurately estimate the uncertainty of the system and adjust the sliding mode control parameters.

[0055] Optionally, the determining of the system initial state and the target equilibrium point, and determining the relationship function between the coefficient of the exponential reaching law and the sliding hyperplane according to the system initial state and the target equilibrium point, are specifically:

[0056] The simplified mathematical model

[0057]

[0058] Further simplified to

[0059]

[0060] The relationship function between the coefficient of the exponential reaching law and the sliding hyperplane is defined as:

[0061]

[0062] Among them, s(x) is the switching function, u eq is equivalent control, k is the reaching law parameter, is the uncertainty of the neural network estimate; u + ≠u - , s(x)=0 is the switching surface. If the system has not reached the sliding surface, the state trajectory can reach the switching surface within a finite time; when the system is on the sliding surface, it can slide along the sliding surface toward the equilibrium point and become asymptotically stable.

[0063] Optionally, the three-layer feedforward neural network selects sigmoid as an activation function.

[0064] Optionally, the final control signal acts on the interconnected dual-machine power system of the SVC, and changes the current flowing through the SVC by controlling the trigger angle of the thyristor, that is, changes the equivalent susceptance of the SVC, thereby controlling the reactive power value injected or absorbed by the SVC into the power system, thereby achieving the goal of regulating the system voltage and suppressing chaotic oscillations of the power system.

[0065] The present invention also proposes a power system chaotic oscillation control system based on a reactive power compensation device, comprising:

[0066] A model building module is used to build a mathematical model of an interconnected dual-machine power system containing SVC based on dynamic equations;

[0067] A model simplification module, used to simplify the mathematical model to obtain a simplified mathematical model;

[0068] The control function calculation module is used to design the switching function according to the state variables of the simplified model, and to design the variable structure sliding mode control law based on the Lyapunov stability theory to obtain the sliding mode control function;

[0069] A relationship function calculation module is used to determine the initial state of the system and the target equilibrium point, and to determine the relationship function between the coefficient of the exponential reaching law and the sliding hyperplane according to the initial state of the system and the target equilibrium point in combination with the switching function and the sliding mode control function;

[0070] A neural network building module is used to build a three-layer feedforward neural network, which is trained based on the relationship function through an operating data set of an interconnected two-machine power system containing an SVC;

[0071] A control signal calculation module is used to feed back the state variables of the interconnected dual-machine power system containing SVC to the sliding mode control function and the neural network in real time. The sliding mode control function calculates the preliminary control signal according to the current state and the switching function, and the neural network calculates the adjustment value of the sliding mode control parameter and the compensation value for the system uncertainty according to the input state variables;

[0072] The chaotic oscillation control module is used to act on the interconnected dual-machine power system of the SVC with the sum of the preliminary control signal, the adjustment value of the sliding mode control parameter and the compensation value for the system uncertainty as the final control signal, so as to guide the system state to converge to a stable state; at the same time, the system state is continuously monitored, and the weights of the neural network and the parameters of the sliding mode control are continuously adjusted according to new data, so that the coefficients of the exponential reaching law and the sliding hyperplane that the neural network is applied to in the sliding mode variable structure are close to each other.

[0073] The beneficial effects of the above technical solution of the present invention are as follows:

[0074] 1. The reactive power demand in the power system is complex and changeable, and is affected by many factors such as load type and load change rate. Neural networks have powerful nonlinear mapping capabilities and can learn and approximate the complex relationship between reactive power and system status, thereby providing more accurate reference information for sliding mode control and improving the accuracy of reactive compensation.

[0075] 2. Sliding mode control itself is robust to system uncertainty and external interference. In reactive power compensation, there may be uncertainties such as voltage fluctuations and harmonic interference in the power grid. Neural networks can estimate these uncertainties in real time, and by combining with sliding mode control, further enhance the resistance of the entire reactive power compensation system to these interferences, ensuring the stability of reactive power compensation.

[0076] 3. In the sliding mode control of reactive power compensation, the design of the sliding surface is crucial to the control performance. The neural network can dynamically adjust the parameters of the sliding surface according to the real-time status of the power system (such as voltage amplitude, phase angle, reactive power, etc.), so that the sliding surface is more adapted to the current working conditions of the system.

[0077] 4. The reaching law parameters, switching gain and other control parameters in the sliding mode control can be optimized through neural networks. The neural network uses learning algorithms (such as gradient descent method) to adjust these control parameters according to the performance indicators in the reactive power compensation process (such as the power factor after compensation, voltage fluctuation range, etc.), thereby effectively suppressing the chattering phenomenon of the sliding mode variable structure control, making the chaotic oscillation control more ideal, and improving the efficiency and stability of reactive power compensation. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] Figure 1 It is a flow chart of the chaotic oscillation control method of the power system based on the reactive power compensation device of the present invention;

[0079] Figure 2 This is an equivalent two-machine system diagram containing SVC of the present invention;

[0080] Figure 3 The SVC of the present invention is attached with a control structure diagram;

[0081] Figure 4 The schematic diagram of the chaotic oscillation control system of the power system based on the reactive power compensation device of the present invention is shown in FIG. DETAILED DESCRIPTION

[0082] In order to make the technical problems, technical solutions and advantages to be solved by the present invention more clear, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0083] like Figure 1 As shown, an embodiment of the present invention provides a method for controlling chaotic oscillation of a power system based on a reactive power compensation device, comprising the following steps:

[0084] Step S1, establishing a mathematical model of an interconnected dual-machine power system including an SVC according to a dynamic equation;

[0085] Step S2, simplifying the mathematical model to obtain a simplified mathematical model;

[0086] Step S3, designing a switching function according to the state variables of the simplified model, and designing a variable structure sliding mode control law based on Lyapunov stability theory to obtain a sliding mode control function;

[0087] Step S4, determining the system initial state and the target equilibrium point, combining the switching function and the sliding mode control function, and determining the relationship function between the coefficient of the exponential reaching law and the sliding hyperplane according to the system initial state and the target equilibrium point;

[0088] Step S5, constructing a three-layer feedforward neural network, and training it based on the relationship function using an operating data set of an interconnected dual-machine power system containing SVC;

[0089] Step S6: The state variables of the interconnected dual-machine power system containing the SVC are fed back to the sliding mode control function and the neural network in real time. The sliding mode control function calculates a preliminary control signal according to the current state and the switching function. The neural network calculates the adjustment value of the sliding mode control parameter and the compensation value for the system uncertainty according to the input state variables.

[0090] Step S7, using the sum of the preliminary control signal, the adjustment value of the sliding mode control parameter and the compensation value for the system uncertainty as the final control signal to act on the interconnected dual-machine power system of the SVC, and guiding the system state to converge to a stable state; at the same time, continuously monitoring the state of the system, and continuously adjusting the weights of the neural network and the parameters of the sliding mode control according to new data, so that the coefficients of the neural network applied to the sliding mode variable structure approximate the relationship function of the exponential reaching law and the sliding hyperplane.

[0091] In this embodiment, a mathematical model of an interconnected dual-machine power system containing SVC is established based on the dynamic equation, specifically:

[0092] (1) Construct the rotor motion equation of the generator

[0093]

[0094] Among them, δ i , P mi , P ei 、M i are the equivalent generator electrical angle, equivalent generator mechanical power, equivalent generator electromagnetic power and inertia time constant respectively; Figure 2 In the figure, G1 and G2 are regional equivalent generators. The equivalent generator uses the classical model and ignores the line and load losses. It is assumed that the generator damping factor can maintain small disturbance stability normally, the generator has excitation and PSS functions, and the prime mover power and transient potential are constants.

[0095] (2) Expanding equation (1) and establishing mathematical equations based on the second-order model of the power system, the mathematical model of the system without SVC can be obtained as follows:

[0096]

[0097] Among them, δ1 and δ2 represent the power angles of the two equivalent generators; ω1 and ω2 represent the rotor angular velocities of the two equivalent generators; ω 10 ,ω 20 represents the rated rotor angular velocity of the two equivalent generators; H1 and H2 represent the rotational inertia of the two equivalent generators; D1 and D2 represent the damping coefficients of the two equivalent generators; P m1 , P m2 Represents the mechanical power of two equivalent generators; P e1 , P e2 represents the electromagnetic power of two equivalent generators, and the electromagnetic power expression of two equivalent generators G1 and G2 is

[0098]

[0099] Where X∑ is the total reactance of the tie line between the two equivalent machines, X∑=X a +Xb +X a X b B SVC , where B SVC is the susceptance of the SVC;

[0100] (3) When the model contains SVC, the closed-loop system composed of SVC and power system is generally equivalent to the first-order inertia system with closed-loop time constant. The mathematical model of SVC is:

[0101]

[0102] Among them, T, u and B SVC0 They are the inertia time constant, control quantity and initial value of susceptance of SVC respectively;

[0103] (4) In an equivalent two-machine system, it is meaningless to discuss the power angle and angular velocity of a certain equivalent generator alone. In order to better demonstrate the stability of the equivalent two-machine system, this paper uses the relative power angle and relative angular velocity as state variables. Let δ = δ1-δ2-δ0, ω = ω1-ω2, δ0 is the initial value of the power angle difference between the two equivalent generators, and substitute equation (3) into equation (2), then equation (2) can be transformed into

[0104]

[0105] (5) Since the damping coefficient of the system cannot be measured accurately, it is defined as an uncertain parameter. In addition, assuming that the damping of the two-machine equivalent system with SVC is uniform, let θ = -D1 / H1 = -D2 / H2, define state variables x1 = δ, x2 = ω to transform the system (5) into

[0106]

[0107] Also defined Process some system parameters as follows

[0108]

[0109] k2=k3(-B SVC +B SVC0 ) (8)

[0110]

[0111] Taking the derivative of x3 and substituting equation (4) into it, we can get

[0112]

[0113] In this embodiment, the mathematical model of step 1 is simplified to obtain a simplified model, specifically:

[0114] Let f(x1,x2)=k1x2cos(x1+δ0)+k2sin(x1+δ0)

[0115] Taking into account the disturbances that may exist in the actual system, these disturbances may be caused by internal perturbations and external disturbances to the system itself, or by the interference of the measurement signal delay of the SVC first-order inertial system. At the same time, taking into account the inaccuracy caused by the simplified means adopted when establishing the model, the interference term w(t) is introduced, and it is assumed that w(t) has an upper bound, that is, |w(t)|≤ρ, in order to improve the accuracy of the model and ensure the robustness of the designed controller, the following simplified mathematical model is obtained

[0116]

[0117] like Figure 3 The figure shows the overall control structure of the SVC including the additional controller. The static VAR compensator (SVC) controls the amount of reactive power injected into or absorbed from the power system through power electronic devices, thereby adjusting the voltage at its terminals and improving the transient stability of the power system. When the system voltage is low, the SVC generates reactive power (SVC capacitive); when the system voltage is high, the SVC absorbs reactive power (SVC inductive). QSVC is the reactive power provided by the SVC, which is expressed as

[0118] Q svc =B svc V 2 svc

[0119] As can be seen from the above formula, reactive power can be regulated by changing the SVC susceptance or changing the SVC voltage, but the SVC voltage is generally the voltage of the node connecting the SVC and the system, which cannot be adjusted freely. Therefore, it is considered to change the SVC susceptance to control reactive power output. The TSC-TCR type SVC can change the current flowing through the SVC by designing an effective controller to control the trigger angle of the thyristor in real time to switch the appropriate capacitor bank and inductance on the secondary side of the coupling transformer, which is equivalent to changing the SVC susceptance and indirectly executing the change of reactive power.

[0120] Figure 3 In, V, V ref are the actual measured voltage and reference voltage at the SVC installation point respectively; α is the SVC trigger angle, α max , α min are the upper and lower limits of the trigger angle, B svc is the SVC susceptance value, B max , B minThe sliding surface selected by the common sliding mode controller in the design process is usually linear. The defect of the linear sliding surface is that the state tracking error cannot converge to zero within a finite time. Therefore, the present invention considers using a special sliding mode control method to design a switching function according to the state variables of the simplified model, specifically:

[0121]

[0122] Among them, s0=cx1+x2, α0 and β0 are both greater than 0, q0 and p0 are positive odd numbers, and C is a normal number.

[0123] In this embodiment, a variable structure sliding mode control law is designed based on Lyapunov stability theory, and a sliding mode control function is obtained, which is specifically:

[0124] The Lyapunov function of sliding mode control is defined as:

[0125]

[0126] Taking the first-order derivative of the Lyapunov function, we get:

[0127] in

[0128]

[0129] In order to make the sliding mode reach the condition The equivalent sliding mode control function is designed as:

[0130]

[0131] Among them, k3 is the convergence rate and C is a positive constant.

[0132] In this embodiment, a three-layer feedforward neural network is constructed and trained through an operation data set of an interconnected dual-machine power system containing SVC to obtain a neural network model, specifically:

[0133] Three-layer feedforward neural network, the input layer has 3 neurons, corresponding to x1, x2, x3 of the simplified mathematical model, the hidden layer has at least 10 neurons, and the output layer has 2 neurons for outputting the adjustment value of the sliding mode control parameter and the compensation value for the system uncertainty respectively;

[0134] The simulation data and some actual operation data of the interconnected dual-machine power system containing SVC are used for training. During the training process, the weights and biases of the neural network are adjusted according to the back propagation algorithm with adaptive learning rate until the output of the neural network meets the set accuracy requirements, that is, it can accurately estimate the uncertainty of the system and adjust the sliding mode control parameters. Among them, the three-layer feedforward neural network selects sigmoid as the activation function.

[0135] In this embodiment, the initial state of the system and the target equilibrium point are determined, and the relationship function between the coefficient of the exponential reaching law and the sliding hyperplane is determined according to the initial state of the system and the target equilibrium point, which is specifically:

[0136] The mathematical model will be simplified

[0137]

[0138] Further simplified to

[0139]

[0140] The relationship function between the coefficient of the exponential reaching law and the sliding hyperplane is defined as:

[0141]

[0142] Among them, s(x) is the switching function, u eq is equivalent control, k is the reaching law parameter, is the uncertainty of the neural network estimate; u + ≠u - , s(x)=0 is the switching surface. If the system has not reached the sliding surface, the state trajectory can reach the switching surface within a finite time; when the system is on the sliding surface, it can slide along the sliding surface toward the equilibrium point and become asymptotically stable.

[0143] In this embodiment, the final control signal acts on the interconnected dual-machine power system of the SVC. For the SVC, the current flowing through the SVC is changed by controlling the trigger angle of the thyristor, that is, the equivalent susceptance of the SVC is changed, and then the reactive power value injected or absorbed by the SVC into the power system is controlled, so as to achieve the purpose of regulating the system voltage and suppressing the chaotic oscillation of the power system.

[0144] In sliding mode control, the uncertainty of the system (such as unmodeled dynamics, external disturbances, etc.) is an important factor affecting the control effect. Neural networks can be used to estimate these uncertainties. For example, for a nonlinear system Where d(x) is the uncertainty part. Neural networks can be trained to approximate d(x).

[0145] When training a neural network, you can use the input and output data of the system. For example, collect the state and control input of the system as the input of the neural network, and use the error between the actual output of the system and the output calculated based on the nominal model as the training target of the neural network.

[0146] There are some key parameters in sliding mode control, such as sliding surface parameters, reaching law parameters, etc. Neural networks can dynamically adjust these parameters according to the performance indicators of the system (such as tracking error, overshoot, etc.). Taking the adjustment of sliding surface parameters as an example, assume that the sliding surface is s = Cx, where C is the parameter matrix. The neural network can continuously update the value of C through the learning algorithm according to the current state and performance of the system, making the design of the sliding surface more reasonable, thereby improving the control performance of the system.

[0147] The present invention first establishes a mathematical model of an interconnected dual-machine power system containing SVC based on the dynamic equation, and simplifies the mathematical model to obtain a simplified mathematical model to determine the state equation and control target of the system. Then, the structure (including the number of layers, the number of neurons, etc.) and parameters (such as weights, biases, etc.) of the neural network are initialized, and the neural network is trained using the operating data of the system. A back propagation algorithm or other advanced training algorithms can be used. During the training process, the error between the actual output and the expected output of the system is used as a loss function, and the parameters of the neural network are continuously adjusted to minimize the loss function. For example, when training a neural network for estimating uncertainty, the error between the actual state change of the system and the state change calculated based on the sliding mode control after estimating the uncertainty is used as a training target, and through multiple iterative training, the neural network can accurately estimate the uncertainty.

[0148] In the actual control process, neural networks and sliding mode control work together. The neural network continuously estimates uncertainty or adjusts sliding mode control parameters, and the sliding mode control calculates the control input based on the output of the neural network and its own control law. For example, in each control cycle, the neural network first estimates the current uncertainty d, and then the sliding mode control calculates the control input based on the estimated uncertainty and the sliding mode control law.

[0149]

[0150] where u eq is an equivalent control, used to calculate a preliminary control signal; k is a reaching law parameter, i.e., an adjustment value of the sliding mode control parameter of the present invention; is the uncertainty estimated by the neural network, i.e., the compensation value of the system uncertainty described in the present invention. The present invention uses the sum of the preliminary control signal, the adjustment value of the sliding mode control parameter and the compensation value for the system uncertainty as the final control signal to act on the interconnected dual-machine power system of the SVC, guiding the system state to converge to a stable state; at the same time, the system state is continuously monitored, and the weights of the neural network and the parameters of the sliding mode control are continuously adjusted according to new data, so that the coefficients of the exponential convergence law and the sliding hyperplane that the neural network is applied to in the sliding mode variable structure are close to the relationship function.

[0151] For SVC, the current flowing through the SVC is changed by controlling the trigger angle of the thyristor, that is, the equivalent susceptance of the SVC is changed, and then the reactive power value injected or absorbed by the SVC into the power system is controlled, so as to achieve the purpose of regulating the system voltage and suppressing the chaotic oscillation of the power system.

[0152] like Figure 4 As shown, the present invention also proposes a power system chaotic oscillation control system based on a reactive power compensation device, comprising:

[0153] A model building module 101 is used to build a mathematical model of an interconnected dual-machine power system including an SVC according to a dynamic equation;

[0154] The model simplification module 102 is used to simplify the mathematical model to obtain a simplified mathematical model;

[0155] A control function calculation module 103 is used to design a switching function according to the state variables of the simplified model, and to design a variable structure sliding mode control law based on Lyapunov stability theory to obtain a sliding mode control function;

[0156] The relationship function calculation module 104 is used to determine the system initial state and the target equilibrium point, and to determine the relationship function between the coefficient of the exponential convergence law and the sliding hyperplane according to the system initial state and the target equilibrium point by combining the switching function and the sliding mode control function;

[0157] A neural network construction module 105 is used to construct a three-layer feedforward neural network, and train it through an operation data set of an interconnected two-machine power system containing SVC according to a relational function;

[0158] A control signal calculation module 106 is used to feed back the state variables of the interconnected dual-machine power system containing the SVC to the sliding mode control function and the neural network in real time, the sliding mode control function calculates the preliminary control signal according to the current state and the switching function, and the neural network calculates the adjustment value of the sliding mode control parameter and the compensation value for the system uncertainty according to the input state variables;

[0159] The chaotic oscillation control module 107 is used to use the sum of the preliminary control signal, the adjustment value of the sliding mode control parameter and the compensation value for the system uncertainty as the final control signal to act on the interconnected dual-machine power system of the SVC, so as to guide the system state to converge to a stable state; at the same time, the system state is continuously monitored, and the weights of the neural network and the parameters of the sliding mode control are continuously adjusted according to new data, so that the coefficients of the neural network applied to the sliding mode variable structure approximate the relationship function between the exponential reaching law and the sliding hyperplane.

[0160] The implementation methods of each module of this system correspond to the above-mentioned chaotic oscillation control method of the power system based on the reactive power compensation device, and will not be repeated here.

[0161] The present invention combines the sliding mode control of the power system with the neural network control, which has the following effects:

[0162] 1. The reactive power demand in the power system is complex and changeable, and is affected by many factors such as load type and load change rate. Neural networks have powerful nonlinear mapping capabilities and can learn and approximate the complex relationship between reactive power and system status, thereby providing more accurate reference information for sliding mode control and improving the accuracy of reactive compensation.

[0163] 2. Sliding mode control itself is robust to system uncertainty and external interference. In reactive power compensation, there may be uncertainties such as voltage fluctuations and harmonic interference in the power grid. Neural networks can estimate these uncertainties in real time, and by combining with sliding mode control, further enhance the resistance of the entire reactive power compensation system to these interferences, ensuring the stability of reactive power compensation.

[0164] 3. In the sliding mode control of reactive power compensation, the design of the sliding surface is crucial to the control performance. The neural network can dynamically adjust the parameters of the sliding surface according to the real-time state of the power system (such as voltage amplitude, phase angle, reactive power, etc.). For example, through an adaptive neural network, the coefficients in the sliding surface equation are modified in real time according to the error between the system state and the desired reactive power compensation target, so that the sliding surface is more adapted to the current working conditions of the system.

[0165] 4. The reaching law parameters, switching gain and other control parameters in the sliding mode control can be optimized through neural networks. The neural network uses learning algorithms (such as gradient descent method) to adjust these control parameters according to the performance indicators in the reactive power compensation process (such as the power factor after compensation, voltage fluctuation range, etc.), thereby effectively suppressing the chattering phenomenon of the sliding mode variable structure control, making the chaotic oscillation control more ideal, and improving the efficiency and stability of reactive power compensation.

[0166] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for controlling chaotic oscillation of a power system based on a reactive power compensation device, characterized in that: The following steps are involved: Step 1: Establish a mathematical model of an interconnected dual-machine power system containing SVC based on dynamic equations; Step 2, simplifying the mathematical model to obtain a simplified mathematical model; Step 3: Design a switching function based on the state variables of the simplified model, and design a variable structure sliding mode control law based on Lyapunov stability theory to obtain a sliding mode control function; Step 4: Determine the initial state of the system and the target equilibrium point, combine the switching function and the sliding mode control function, and determine the relationship function between the coefficient of the exponential reaching law and the sliding hyperplane according to the initial state of the system and the target equilibrium point; Step 5: construct a three-layer feedforward neural network, and train it based on the relationship function using an operating data set of an interconnected dual-machine power system containing SVC; Step 6: The state variables of the interconnected dual-machine power system containing the SVC are fed back to the sliding mode control function and the neural network in real time. The sliding mode control function calculates a preliminary control signal according to the current state and the switching function. The neural network calculates the adjustment value of the sliding mode control parameter and the compensation value for the system uncertainty according to the input state variables. Step 7: The sum of the preliminary control signal, the adjustment value of the sliding mode control parameter and the compensation value for the system uncertainty is used as the final control signal to act on the interconnected dual-machine power system of the SVC to guide the system state to converge to a stable state; at the same time, the state of the system is continuously monitored, and the weights of the neural network and the parameters of the sliding mode control are continuously adjusted according to new data, so that the neural network is applied to the sliding mode variable structure to approximate the relationship function between the coefficients of the exponential reaching law and the sliding hyperplane.

2. The method for controlling chaotic oscillation of a power system based on a reactive power compensation device according to claim 1, characterized in that: The mathematical model of the interconnected dual-machine power system containing SVC is established according to the dynamic equation, specifically: (1) Construct the rotor motion equation of the generator Among them, δ i , P mi , P ei 、M i are the equivalent generator electrical angle, equivalent generator mechanical power, equivalent generator electromagnetic power and inertia time constant respectively; (2) Expanding equation (1) and establishing mathematical equations based on the second-order model of the power system, the mathematical model of the system without SVC can be obtained as follows: Among them, δ1 and δ2 represent the power angles of the two equivalent generators; ω1 and ω2 represent the rotor angular velocities of the two equivalent generators; ω 10 ,ω 20 represents the rated rotor angular velocity of the two equivalent generators; H1 and H2 represent the rotational inertia of the two equivalent generators; D1 and D2 represent the damping coefficients of the two equivalent generators; P m1 , P m2 Represents the mechanical power of two equivalent generators; P e1 , P e2 represents the electromagnetic power of two equivalent generators, and the electromagnetic power expression of two equivalent generators G1 and G2 is Where X∑ is the total reactance of the tie line between the two equivalent machines, X∑=X a +X b +X a X b B SVC , where B SVC is the susceptance of the SVC; (3) When the model contains SVC, the closed-loop system composed of SVC and power system is generally equivalent to the first-order inertia system with closed-loop time constant. The mathematical model of SVC is: Among them, T, u and B SVC0 They are the inertia time constant, control quantity and initial value of susceptance of SVC respectively; (4) Using relative power angle and relative angular velocity as state variables, let δ = δ1-δ2-δ0, ω=ω1-ω2, δ0 is the initial value of the power angle difference between the two equal generators, and substitute equation (3) into equation (2), Then we can transform formula (2) into (5) Let θ = -D1 / H1 = -D2 / H2, define the state variables x1 = δ, x2 = ω, and transform the system (5) into Also defined Process some system parameters as follows k2=k3(-B SVC +B SVC0 ) (8) Taking the derivative of x3 and substituting equation (4) into it, we can get 3. The method for controlling chaotic oscillation of a power system based on a reactive power compensation device according to claim 2, characterized in that: The mathematical model of step 1 is simplified to obtain a simplified model, which is specifically: Let f(x1,x2)=k1x2cos(x1+δ0)+k2sin(x1+δ0) And introduce the interference term w(t), and assume that w(t) has an upper bound, that is, |w(t)|≤ρ, and get the following simplified mathematical model 4. The method for controlling chaotic oscillation of a power system based on a reactive power compensation device according to claim 1, characterized in that: The switching function is designed according to the state variables of the simplified model, specifically: Among them, s0=cx1+x2, α0 and β0 are both greater than 0, q0 and p0 are positive odd numbers, and C is a normal number.

5. The method for controlling chaotic oscillation of a power system based on a reactive power compensation device according to claim 4, characterized in that: The variable structure sliding mode control law is designed based on Lyapunov stability theory, and the sliding mode control function is obtained, which is specifically: The Lyapunov function of sliding mode control is defined as: Taking the first-order derivative of the Lyapunov function, we get: in In order to make the sliding mode reach the condition The equivalent sliding mode control function is designed as: Among them, k3 is the convergence rate and C is a positive constant.

6. The method for controlling chaotic oscillation of a power system based on a reactive power compensation device according to claim 1, characterized in that: The three-layer feedforward neural network is constructed and trained by the operation data set of the interconnected dual-machine power system containing SVC to obtain a neural network model, which is specifically: The three-layer feedforward neural network has 3 neurons in the input layer, corresponding to x1, x2, and x3 of the simplified mathematical model, at least 10 neurons in the hidden layer, and 2 neurons in the output layer, respectively used to output the adjustment value of the sliding mode control parameter and the compensation value for the system uncertainty; The simulation data of the interconnected dual-machine power system containing SVC and some actual operation data are used for training. During the training process, the weights and biases of the neural network are adjusted according to the back-propagation algorithm with adaptive learning rate until the output of the neural network meets the set accuracy requirements, that is, it can accurately estimate the uncertainty of the system and adjust the sliding mode control parameters.

7. The method for controlling chaotic oscillation of a power system based on a reactive power compensation device according to claim 1, characterized in that: The determination of the system initial state and the target equilibrium point, and the relationship function between the coefficient of the exponential reaching law and the sliding hyperplane according to the system initial state and the target equilibrium point, are specifically: The simplified mathematical model Further simplified to The relationship function between the coefficient of the exponential reaching law and the sliding hyperplane is defined as: Among them, s(x) is the switching function, u eq is equivalent control, k is the reaching law parameter, is the uncertainty of the neural network estimate; u + ≠u - , s(x)=0 is the switching surface. If the system has not reached the sliding surface, the state trajectory can reach the switching surface within a finite time; when the system is on the sliding surface, it can slide along the sliding surface toward the equilibrium point and become asymptotically stable.

8. The method for controlling chaotic oscillation of a power system based on a reactive power compensation device according to claim 1, characterized in that: The three-layer feedforward neural network selects sigmoid as the activation function.

9. The method for controlling chaotic oscillation of a power system based on a reactive power compensation device according to claim 1, characterized in that: The final control signal acts on the interconnected dual-machine power system of the SVC. For the SVC, the current flowing through the SVC is changed by controlling the trigger angle of the thyristor, that is, the equivalent susceptance of the SVC is changed, and then the reactive power value injected or absorbed by the SVC into the power system is controlled, so as to achieve the purpose of regulating the system voltage and suppressing the chaotic oscillation of the power system.

10. A power system chaotic oscillation control system based on a reactive power compensation device, characterized in that: include: A model building module is used to build a mathematical model of an interconnected dual-machine power system containing SVC based on dynamic equations; A model simplification module, used to simplify the mathematical model to obtain a simplified mathematical model; The control function calculation module is used to design the switching function according to the state variables of the simplified model, and to design the variable structure sliding mode control law based on the Lyapunov stability theory to obtain the sliding mode control function; A relationship function calculation module is used to determine the initial state of the system and the target equilibrium point, and to determine the relationship function between the coefficient of the exponential reaching law and the sliding hyperplane according to the initial state of the system and the target equilibrium point in combination with the switching function and the sliding mode control function; A neural network building module is used to build a three-layer feedforward neural network, which is trained based on the relationship function through an operating data set of an interconnected two-machine power system containing an SVC; A control signal calculation module is used to feed back the state variables of the interconnected dual-machine power system containing SVC to the sliding mode control function and the neural network in real time. The sliding mode control function calculates the preliminary control signal according to the current state and the switching function, and the neural network calculates the adjustment value of the sliding mode control parameter and the compensation value for the system uncertainty according to the input state variables; The chaotic oscillation control module is used to act on the interconnected dual-machine power system of the SVC with the sum of the preliminary control signal, the adjustment value of the sliding mode control parameter and the compensation value for the system uncertainty as the final control signal, so as to guide the system state to converge to a stable state; at the same time, the system state is continuously monitored, and the weights of the neural network and the parameters of the sliding mode control are continuously adjusted according to new data, so that the coefficients of the exponential reaching law and the sliding hyperplane that the neural network is applied to in the sliding mode variable structure are close to each other.

Images