Reactive voltage dynamic optimization method considering wind-thermal power generation through direct current transmission

Abstract

The invention provides a reactive voltage dynamic optimization method considering wind-thermal power generation through direct current transmission, and the method comprises the following steps: S1, building a reactive voltage optimization model of a wind power injection LCC-HVDC alternating-current and direct-current system, and setting constraint conditions; s2, measuring the transient overvoltage suppression effect of the operation parameters through a transient overvoltage suppression index, evaluating and screening key optimization parameters, and constructing a parameter dynamic optimization control system structure; and S3, carrying out calculation learning by using a deep fuzzy neural network, dynamically generating a STATCOM parameter optimization instruction, realizing dynamic reactive voltage optimization, and analyzing an optimization result. The invention provides a reactive voltage dynamic optimization method considering wind-thermal power generation and direct-current transmission, which takes key operation parameters considering wind power injection and a high-voltage direct-current transmission system as input quantities to carry out learning calculation, and carries out dynamic optimization on STATCOM parameters, so that the reactive support level of the STATCOM is dynamically adjusted under the transient extreme condition of the system, and the system stability is improved. The voltage instability condition is improved, and the system reliability is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of voltage control of new energy AC / DC systems, and in particular relates to a method for dynamically optimizing reactive voltage taking into account wind-thermal power generation via DC transmission. Background Art

[0002] With the continued development of renewable energy generation and the large-scale application of long-distance direct current (DC) transmission technology in my country, the entire power grid faces frequent voltage overshoots and weak reactive power support, posing significant challenges to the dispatching, operation, and control of power systems. Static synchronous compensators (STATCOMs), with their advantages such as a small footprint, fast regulation speed, and strong bidirectional dynamic regulation capabilities, play a vital role in optimizing reactive power and voltage in power systems. However, how to more fully and flexibly unleash the STATCOM's inherent dynamic reactive power regulation capabilities, further explore its dynamic regulation characteristics, and dynamically optimize its control parameters is a pressing research topic.

[0003] Currently, domestic and international scholars focusing on STATCOM optimization control primarily focus on control strategy design and more traditional parameter optimization. Optimization algorithms primarily rely on trial-and-error methods, approximate linearization methods, and older algorithms such as fuzzy adaptive and particle swarm optimization. With the increasing penetration of renewable energy and the development of power grids influenced by multiple factors such as high-voltage direct current transmission, the complexity of reactive power and voltage optimization is gradually increasing, and the requirements for transient stability of power systems are becoming increasingly stringent. Currently, with the rapid development of emerging technologies such as big data, deep learning, and digital twins, real-time optimization control that deeply integrates virtual-reality interactions with artificial intelligence algorithms will become an inevitable trend in the development of power systems.

[0004] Therefore, the patent of this invention proposes a deep fuzzy neural network algorithm, which combines the network structure and learning algorithm of the deep neural network with easy-to-understand fuzzy reasoning language rules, and deeply explains how to use the error back propagation algorithm in this deep fuzzy network for parameter learning, to optimize the key control parameters of STATCOM in real time, improve the dynamic reactive compensation effect in the new energy AC and DC power system, release the reactive regulation potential under transient extreme conditions, and improve the transient characteristics of the system voltage. Summary of the Invention

[0005] The technical problem to be solved by the present invention is that a dynamic optimization method for reactive voltage taking into account wind and thermal power generation via DC transmission is proposed through a deep fuzzy neural network algorithm. The method uses key operating parameters of wind power injection and high-voltage DC transmission systems as input for learning calculations, and dynamically optimizes STATCOM parameters, thereby dynamically adjusting the reactive support level of STATCOM under extreme transient conditions of the system, improving voltage instability, and enhancing system reliability.

[0006] The present invention specifically provides a method for dynamically optimizing reactive power and voltage by taking into account wind-thermal power generation via direct current transmission, comprising the following steps:

[0007] S1: Build a reactive power and voltage optimization model for the LCC-HVDC AC / DC system with wind power injection and set constraints;

[0008] S2: Based on the reactive voltage optimization model established in step S1, the transient overvoltage suppression effect of the operating parameters is measured by the transient overvoltage suppression index, key optimization parameters are evaluated and selected, and a parameter dynamic optimization control system structure is constructed;

[0009] S3: Based on the key operating parameters selected in step S2, a deep fuzzy neural network is used for computational learning to dynamically generate STATCOM parameter optimization instructions, achieve dynamic reactive power and voltage optimization, and analyze the optimization results.

[0010] Preferably, the equation constraint condition of the reactive power voltage optimization model in step S1 is:

[0011]

[0012] Where, P Gi and Q Gi is the active and reactive power of the generator; P Li and Q Li is the active and reactive power of the node load; P idc and Q idc is the active and reactive power of the converter station node; P wi and Q wi is the active and reactive power injected into wind power; Q STATCOM,i is the reactive compensation of STATCOM; U i and U j represents the voltage amplitude of nodes i and j; G ij 、B ij and θ ij are the conductance, susceptance and node voltage phase angle difference between nodes i and j respectively, and n is the number of system nodes.

[0013] Preferably, the inequality constraints of the reactive power voltage optimization model in step S1 mainly include upper and lower limits of control variables, system generator output and node voltage constraints, which are specifically expressed as follows:

[0014]

[0015] Where: U imin and U imax Indicates the upper and lower limits of the node i voltage; Q Gimin and Q GimaxIndicates the upper and lower limits of the generator reactive output; Q wimin and Q wimax Indicates the upper and lower limits of wind power reactive power output; Q wimin and Q wimax Indicates the upper and lower limits of wind power reactive power output; Q STATCOM,imax Indicates the upper limit of STATCOM capacity; U idc , I idc and P idc Respectively represent the voltage, current and active power of the DC node; U idcmin and U idcmax Indicates the upper and lower limits of the DC node voltage; I idcmin and I idcmax Indicates the upper and lower limits of DC node current; P idcmin and P idcmax Indicates the upper and lower limits of the DC node active power.

[0016] Preferably, the step S1 uses the voltage fluctuation of the common connection point as the objective function to measure the stability of the system voltage, and the calculation formula is:

[0017]

[0018] Where: U pcc is the actual voltage at the node common connection point, is the expected voltage at the common connection point.

[0019] Preferably, the step S2 measures the severity of the transient overvoltage of the node under transient conditions by using a transient voltage over-limit index, which is expressed as:

[0020]

[0021] Where: U(t) is the instantaneous voltage value of the sending end busbar under transient state, U s (t) is the voltage stability limit of the sending-end busbar, t k is the time sequence of the integral discrete points arranged in the direction of the integral interval, Δt k N is the width corresponding to each discrete point. K The total number of discrete points in the calculation time period.

[0022] Preferably, the step S2 uses the above-mentioned transient voltage limit-exceeding index to formulate a transient overvoltage suppression index, which is used to judge the suppression ability of the transient overvoltage by adjusting different operating parameters, so as to select key operating parameters for optimization. The transient overvoltage suppression index calculation formula is:

[0023]

[0024] Where: K *is the initial value of the operating parameter, and ΔK is the change in the adjusted operating parameter.

[0025] Preferably, the parameter dynamic optimization control system structure in step S2 is:

[0026] Deep fuzzy neural controllers A and B are used to update and optimize the PI controller parameters of the outer and inner loops of the STATCOM in real time. The input of deep fuzzy neural controller A is the error between the rated value and the measured value of the line voltage at the grid connection point, the error change rate and the wind speed, and the output is the optimization instruction ΔK of the PI controller. pA , ΔK iA The input of deep fuzzy neural controller B is the error between the reference value and the measured value of the q-axis current output by the outer loop, the error change rate and the wind speed, and the output is the PI controller optimization instruction ΔK pB , ΔK iB , the proportional gain K of the PI controller p and integral gain K i Updated at each sampling moment according to formula (6).

[0027]

[0028] Where: and They represent the proportional and integral empirical parameters of the STATCOM outer and inner loop PI controllers, ΔK p and ΔK i They respectively represent the real-time generated dynamic optimization instructions for the proportional and integral coefficients.

[0029] Preferably, the deep fuzzy neural network in step S3 adopts the error back propagation algorithm to update the parameters. First, define S h is the hidden layer of the network, and the input x=[x1,x2,…,x d ,…,x D ], the network parameters involved are:

[0030]

[0031] Where: θ h is the parameter matrix of fuzzy membership of input condition variables in fuzzy rules; is the input to the hidden layer S h The parameter of the membership degree of the d-th conditional clause of the f-th rule on h is the coefficient matrix of the conclusion part of the fuzzy rule; is the parameter of the membership degree of the dth conditional clause of the fth rule in the conclusion of the fuzzy rule; μ h is the hidden layer S h The membership matrix corresponding to the node, is the hidden layer node S h The membership value of the d-th conditional clause of the corresponding r-th rule, where r = 1, 2, ..., R, D represents the number of inputs to the hidden layer, and F represents the number of inputs to the hidden layer S h The number of fuzzy rules, R represents the number of hidden layer nodes S h The number of fuzzy rules.

[0032] Preferably, the step S3 of calculating the output values ​​of all hidden layer nodes in the network first requires determining the membership weights of the rule conditions. Since the condition part in the fuzzy rule is composed of multiple conditional clauses, it is necessary to use the t operator to merge the membership values ​​of each clause input variable, which is specifically expressed as:

[0033]

[0034] Where: is the hidden layer node S h The weight of the corresponding rth rule, D represents the number of inputs.

[0035] Preferably, the output of the nodes in the network in step S3 is calculated based on the conclusion part of the fuzzy rules, and the r-dimensional vector composed of the outputs of all rules is:

[0036]

[0037] Where: is the hidden layer node S h The output of the corresponding rule r, v h is the hidden layer node S h The corresponding output vector matrix of all fuzzy rules, is the hidden layer node S h The coefficient of the dth conditional clause of the rth rule corresponding to the conclusion of the fuzzy rule.

[0038] Preferably, in step S3, the network hidden layer node S h The overall output can be expressed as the weighted average sum of the weights and node outputs:

[0039]

[0040] Preferably, after obtaining the outputs of all nodes in the network, step S3 defines the loss function of the entire network as the objective function. The minimum loss function is the goal of the entire network parameter learning, which is defined as:

[0041]

[0042] Where: e is the output error, y is the output value of the network, y dis the target value, J is the loss function, e (n) is the error of the nth group of samples, and N is the total number of samples.

[0043] Preferably, step S3 calculates the loss according to the overall loss function in the network, and then propagates the error of the output layer node back to the entire network, and calculates the gradient of the loss function on each network parameter, and finally updates all parameters in the network by gradient descent. The calculation method of the gradient of the loss function on the output layer node is:

[0044]

[0045] Where, is the coefficient of the hth conditional clause of the rth rule in the conclusion part of the fuzzy rule corresponding to the output layer node O, is the output corresponding to the rth rule of the output layer node O, is the weight of the rth rule of the output layer node O.

[0046] Preferably, the calculation of the gradient of the hidden layer parameters in step S3 is similar to the calculation of the gradient of the output layer y. After the gradients corresponding to all parameters in the network are calculated, the parameters can be updated synchronously, which is specifically expressed as follows:

[0047]

[0048] Where: ψ is all network parameters, where ψ new and ψ old Represent the network parameters before and after the update respectively; α is the rate of parameter learning.

[0049] Preferably, after the STATCOM parameters are dynamically optimized by the deep fuzzy neural network algorithm in step S3, the reactive power support effect under different transient disturbances can be improved, and voltage fluctuations and drops can be suppressed. The optimization effect evaluation index is defined as:

[0050]

[0051] E va =γ·TRSA+VFQI (20)

[0052] Where TRSA is the transient reactive power support capability index, ΔU d is the amplitude difference of voltage fluctuation under disturbance; U N is the rated voltage; Δt min Δt is the time difference from the disappearance of the disturbance to the voltage recovery to 0.95 pu; maxis the time difference from the disappearance of disturbance to the restoration of voltage stability; β is the proportional weight; VFQI is the voltage fluctuation qualification index, n is the number of fluctuations within a certain fluctuation time T, n ex E is the number of times the voltage fluctuates by more than 1% per unit time. va is the final optimization effect evaluation index, and γ is the weight coefficient.

[0053] Compared with the prior art, the present invention has the following significant advantages:

[0054] This paper proposes a dynamic parameter optimization method for static synchronous compensators based on wind farms and high-voltage direct current transmission systems, builds a reactive power and voltage optimization model for new energy AC / DC systems taking into account wind farms, LCC-HVDC and STATCOM, uses TS-type fuzzy systems as basic neural units for multi-level combination, and dynamically optimizes STATCOM parameters through a deep fuzzy neural network optimization algorithm with repeated iterative updates through backpropagation, thereby improving the accuracy and sensitivity of learning calculations, fully releasing the dynamic reactive power support potential of STATCOM, and suppressing voltage fluctuations under disturbances. This is of great significance for enhancing the system voltage transient recovery capability and improving the system transient response characteristics. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 This is a flow chart of a method for dynamically optimizing reactive power and voltage by taking into account wind and thermal power generation via direct current transmission according to an embodiment of the present invention;

[0056] Figure 2 1 is a structural diagram of a STATCOM parameter dynamic optimization control system based on a deep fuzzy neural network according to an embodiment of the present invention;

[0057] Figure 3 This is a simplified structural diagram of a deep fuzzy neural network according to an embodiment of the present invention;

[0058] Figure 4 This is a diagram of the deep fuzzy neural network learning calculation process according to an embodiment of the present invention;

[0059] Figure 5 is a topological diagram of a simulation system according to an embodiment of the present invention;

[0060] Figure 6 3. This is a comparison diagram of the bus voltage response curves at the sending end before and after optimization during wind speed disturbance according to an embodiment of the present invention;

[0061] Figure 7 3. This is a comparison diagram of the receiving-end bus voltage response curves before and after optimization during wind speed disturbance according to an embodiment of the present invention;

[0062] Figure 8 3 is a comparison diagram of STATCOM reactive power output curves before and after optimization during wind speed disturbance according to an embodiment of the present invention. DETAILED DESCRIPTION

[0063] The following describes in detail a specific implementation method of a method for calculating the uplift bearing capacity formula of a spiral anchor in a revised specification of the present invention in conjunction with the accompanying drawings.

[0064] The following describes in detail a specific embodiment of a method for dynamically optimizing reactive power and voltage taking into account wind-thermal power generation via direct current transmission according to the present invention with reference to the accompanying drawings.

[0065] like Figure 1 As shown, the present invention provides a technical solution of a specific embodiment:

[0066] Step 1: Build a reactive power and voltage optimization model for the LCC-HVDC AC / DC system with wind power injection, set the equality and inequality constraints of the model, and use the voltage fluctuation at the common connection point as the objective function to measure the stability of the system voltage.

[0067] Step 1.1: When optimizing control parameters, the power system performance constraints such as power balance are required to maintain normal and stable system operation. The equation constraints of the reactive power and voltage optimization model are as follows:

[0068]

[0069] In formula (1), P Gi and Q Gi is the active and reactive power of the generator; P Li and Q Li is the active and reactive power of the node load; P idc and Q idc is the active and reactive power of the converter station node; P wi and Q wi is the active and reactive power injected into wind power; Q STATCOM,i is the reactive compensation of STATCOM; U i and U j represents the voltage amplitude of nodes i and j; G ij 、B ij and θ ij are the conductance, susceptance and node voltage phase angle difference between nodes i and j respectively; n is the number of system nodes.

[0070] Step 1.2: To ensure the safe and stable operation of the power system, all variables must be within the specified range. The inequality constraints of the reactive power and voltage optimization model mainly include the upper and lower limits of the control variables, the system generator output, and the node voltage constraints, which are specifically expressed as:

[0071]

[0072] In formula (2), U imin and U imaxIndicates the upper and lower limits of the node i voltage; Q Gimin and Q Gimax Indicates the upper and lower limits of the generator reactive output; Q wimin and Q wimax Indicates the upper and lower limits of wind power reactive power output; Q STATCOM,imax Indicates the upper limit of STATCOM capacity; U idcmin and U idcmax Indicates the upper and lower limits of the DC node voltage; I idcmin and I idcmax Indicates the upper and lower limits of DC node current; P idcmin and P idcmax Indicates the upper and lower limits of the DC node active power; U idc , I idc and P idc Represent the voltage, current and active power of the DC node respectively.

[0073] Step 1.3: Use the voltage fluctuation at the common connection point as the objective function to measure the stability of the system voltage. The calculation formula is:

[0074]

[0075] In formula (3), U pcc is the actual voltage at the node common connection point, is the expected voltage at the common connection point.

[0076] Step 2: To ensure the effectiveness and sensitivity of parameter optimization, based on the reactive power voltage optimization model established in Step 1, a transient overvoltage suppression index is set to measure the transient overvoltage suppression effect of each operating parameter, evaluate and screen key optimization parameters, and build a parameter dynamic optimization control system structure;

[0077] Step 2.1: To establish the evaluation indicators for key operating parameters, first set the transient voltage limit index to measure the severity of the node transient overvoltage under transient conditions, which is expressed as:

[0078]

[0079] In formula (4), U(t) is the instantaneous voltage value of the sending end busbar under transient state, U s (t) is the voltage stability limit of the sending-end busbar, t k is the time sequence of the integral discrete points arranged in the direction of the integral interval, Δt k N is the width corresponding to each discrete point. K The total number of discrete points in the calculation time period.

[0080] Step 2.2: Use the above transient voltage limit index to further develop the transient overvoltage suppression index, which is used to evaluate the ability of adjusting different operating parameters to suppress transient overvoltage, so as to select key operating parameters for optimization. The transient overvoltage suppression index calculation formula is:

[0081]

[0082] In formula (5), K * is the initial value of the operating parameter, and ΔK is the change in the adjusted operating parameter.

[0083] like Figure 2 As shown, step 2.3: According to the evaluation results of the above steps, the error between the rated value and the measured value of the line voltage at the grid connection point, the error and error change rate, the error between the reference value and the measured value of the q-axis current output by the outer loop, and the wind speed are selected as the input of the deep fuzzy neural network for learning calculation. Based on this, the structure of the dynamic optimization control system of STATCOM parameters is established as follows:

[0084] Deep fuzzy neural controllers A and B are used to update and optimize the PI controller parameters of the outer and inner loops of the STATCOM in real time. The input of deep fuzzy neural controller A is the error between the rated value and the measured value of the line voltage at the grid connection point, the error change rate and the wind speed, and the output is the optimization instruction ΔK of the PI controller. pA , ΔK iA The input of deep fuzzy neural controller B is the error between the reference value and the measured value of the q-axis current output by the outer loop, the error change rate and the wind speed, and the output is the PI controller optimization instruction ΔK pB , ΔK iB , the proportional gain K of the PI controller p and integral gain K i Updated at each sampling moment according to formula (6).

[0085]

[0086] In formula (6), and They represent the proportional and integral empirical parameters of the STATCOM outer and inner loop PI controllers, ΔK p and ΔK i They respectively represent the real-time generated dynamic optimization instructions for the proportional and integral coefficients.

[0087] like Figure 3 and 4As shown in Figure 3, step 3: Based on the key operating parameters and parameter dynamic optimization control system structure screened out in step 2, deep fuzzy neural networks are used in the two controllers to calculate and learn the input quantities, dynamically generate STATCOM parameter optimization instructions and input them into the PI controller to achieve dynamic reactive power and voltage optimization, and analyze the optimization results.

[0088] Step 3.1: Deep fuzzy neural network uses error back propagation algorithm to update parameters. First, define S h is the hidden layer of the network, and the input x=[x1,x2,…,x d ,…,x D ], the network parameters involved are:

[0089]

[0090]

[0091] In formulas (7), (8), and (9): θ h is the parameter matrix of fuzzy membership of input condition variables in fuzzy rules; is the input to the hidden layer S h The parameter of the membership degree of the d-th conditional clause of the f-th rule on h is the coefficient matrix of the conclusion part of the fuzzy rule; is the coefficient of the dth conditional clause of the fth rule in the conclusion part of the fuzzy rule; μ h is the hidden layer S h The membership matrix corresponding to the node, is the hidden layer node S h The membership value of the d-th conditional clause of the corresponding r-th rule, where r = 1, 2, ..., R, D represents the number of inputs to the hidden layer, and F represents the number of inputs to the hidden layer S h The number of fuzzy rules, R represents the number of hidden layer nodes S h The number of fuzzy rules.

[0092] Step 3.2: Calculate the output values ​​of all hidden layer nodes in the network. First, we need to determine the membership weights of the rule conditions. Since the condition part of the fuzzy rule consists of multiple conditional clauses, we need to use the t operator to merge the membership values ​​of each clause input variable. The specific expression is:

[0093]

[0094] In formula (10), is the hidden layer node S h The weight of the corresponding rth rule, D represents the number of inputs.

[0095] Step 3.3: The output of the nodes in the network is calculated based on the conclusion part of the fuzzy rules. The r-dimensional vector composed of the output of all rules is:

[0096]

[0097] In formula (11), is the hidden layer node S h The output of the corresponding rule r, v h is the hidden layer node S h The corresponding output vector matrix of all fuzzy rules, is the hidden layer node S h The coefficient of the dth conditional clause of the rth rule corresponding to the conclusion of the fuzzy rule.

[0098] Step 3.4: Network hidden layer node S h The overall output can be expressed as the weighted average sum of the weights and node outputs:

[0099]

[0100] Step 3.5: After obtaining the output of all nodes in the network, define the loss function of the entire network as the objective function. The minimum loss function is the goal of the entire network parameter learning, which is defined as:

[0101]

[0102] In formula (13), e is the output error, y is the output value of the network, and y d is the target value, J is the loss function, e (n) is the error of the nth group of samples, and N is the total number of samples.

[0103] Step 3.6: Calculate the loss according to the overall loss function of the network, and then propagate the error of the output layer node back to the entire network, and calculate the gradient of the loss function on each network parameter. Finally, update all parameters in the network through the gradient descent method. The calculation method of the gradient of the loss function on the output layer node is:

[0104]

[0105] In formula (15)(16), is the coefficient of the hth conditional clause of the rth rule in the conclusion part of the fuzzy rule corresponding to the output layer node O, is the output corresponding to the rth rule of the output layer node O, is the weight of the rth rule of the output layer node O.

[0106] Step 3.7: The calculation of the gradient of the hidden layer parameters is similar to the calculation of the gradient of the output layer y. Once the gradients corresponding to all parameters in the network are calculated, the parameters can be updated synchronously. Specifically, it is expressed as:

[0107]

[0108] In formula (17), ψ is all network parameters, where ψ new and ψ old Represent the network parameters before and after the update respectively; α is the rate of parameter learning.

[0109] Step 3.8: After the STATCOM parameters are dynamically optimized using the deep fuzzy neural network algorithm, the reactive power support effect under different transient disturbances can be improved, and voltage fluctuations and drops can be suppressed. The optimization effect evaluation index is defined as:

[0110]

[0111] E va =γ·TRSA+VFQI (20)

[0112] In formulas (18), (19), and (20), TRSA is the transient reactive power support capability index, ΔU d is the amplitude difference of voltage fluctuation under disturbance; U N is the rated voltage; Δt min Δt is the time difference from the disappearance of the disturbance to the voltage recovery to 0.95 pu; max is the time difference from the disappearance of disturbance to the restoration of voltage stability; β is the proportional weight; VFQI is the voltage fluctuation qualification index, n is the number of fluctuations within a certain fluctuation time T, n ex E is the number of times the voltage fluctuates by more than 1% per unit time. va is the final optimization effect evaluation index, and γ is the weight coefficient.

[0113] like Figure 5As shown, a simulation model was established for transmitting power from a doubly-fed wind turbine and synchronous motor to the grid via LCC-HVDC. A STATCOM was connected to the sending bus to provide rapid reactive power support, smooth wind power fluctuations, and regulate transient voltage. Simultaneously, a deep fuzzy neural network controller was developed based on MATLAB, with Fortran used for input and output programming. The interface between PSCAD and MATLAB enabled real-time interaction between the STATCOM PI controller parameters and the deep fuzzy neural network gain control instructions in the simulation model, achieving dynamic parameter optimization. The simulation model uses a 345kV AC system at the sending end, connected to synchronous generators and doubly-fed wind turbines. The doubly-fed wind farm consists of 90 5MW wind turbines with a total rated output of 450MW. The synchronous generators consist of three 333MVA synchronous generators connected to the sending busbar via an 18 / 345kV step-up transformer. The STATCOM's maximum reactive power compensation is set at 400MVA, using constant voltage control. The LCC-HVDC model uses a bipolar 12-pulse converter. The DC line is approximately 1000km long, with a rated voltage of ±500kV and a single-pole rated power of 500MW.

[0114] like Figure 6 and Figure 7 As shown in the figure, there is a comparison of the voltage response curves of the sending and receiving busbars before and after deep fuzzy neural network optimization. It can be seen from the figure that when wind speed disturbance occurs, after the STATCOM parameters are optimized, the voltage fluctuation amplitude at both ends is significantly improved, and the voltage recovery speed is significantly shortened. This reactive voltage optimization method has significant advantages in the system transient voltage regulation effect and response speed.

[0115] like Figure 8 As shown in the figure, the comparison of STATCOM reactive power output curves before and after deep fuzzy neural network optimization is shown. It can be seen from the figure that after parameter optimization, the reactive power regulation potential and flexibility of STATCOM are released to a greater extent, and the overall voltage and power quality and transient response characteristics of the system are improved.

[0116] Finally, it should be noted that the above embodiments are merely illustrative of the technical solutions of the present invention and are not intended to limit the present invention. Those skilled in the art will appreciate that they may modify or substitute equivalents for the specific embodiments of the present invention, provided that such modifications or variations are within the scope of protection of the pending claims.

Claims

1. A method for dynamic optimization of reactive power and voltage taking into account wind and thermal power generation via direct current transmission, characterized in that: The reactive voltage dynamic optimization comprises the following steps: S1: Build a reactive power and voltage optimization model for the LCC-HVDC AC / DC system with wind power injection and set constraints; S2: Based on the reactive voltage optimization model established in step S1, the transient overvoltage suppression effect of the operating parameters is measured by the transient overvoltage suppression index, key optimization parameters are evaluated and selected, and a parameter dynamic optimization control system structure is constructed; S3: Based on the key operating parameters selected in step S2, a deep fuzzy neural network is used for computational learning to dynamically generate STATCOM parameter optimization instructions, achieve dynamic reactive power and voltage optimization, and analyze the optimization results.

2. The method for dynamic optimization of reactive power and voltage taking into account wind and thermal power generation via direct current transmission according to claim 1, characterized in that: The equation constraint of the reactive power voltage optimization model in step S1 is: Where, P Gi and Q Gi is the active and reactive power of the generator; P Li and Q Li is the active and reactive power of the node load; P idc and Q idc is the active and reactive power of the converter station node; P wi and Q wi is the active and reactive power injected into wind power; Q STATCOM,i is the reactive compensation of STATCOM; U i and U j represents the voltage amplitude of nodes i and j; G ij 、B ij and θ ij are the conductance, susceptance and node voltage phase angle difference between nodes i and j respectively; n is the number of system nodes; The inequality constraints of the reactive power and voltage optimization model mainly include the upper and lower limits of the control variables, the system generator output, and the node voltage constraints, which can be specifically expressed as follows: Where: U imin and U imax Indicates the upper and lower limits of the node i voltage; Q Gimin and Q Gimax Indicates the upper and lower limits of the generator reactive output; Q wimin and Q wimax Indicates the upper and lower limits of wind power reactive power output; Q STATCOM,imax Indicates the upper limit of STATCOM capacity; U idcmin and U idcmax Indicates the upper and lower limits of the DC node voltage; I idcmin and I idcmax Indicates the upper and lower limits of DC node current; P idcmin and P idcmax Indicates the upper and lower limits of the DC node active power; U idc , I idc and P idc Represent the voltage, current and active power of the DC node respectively.

3. The method for dynamic optimization of reactive power and voltage taking into account wind and thermal power generation via direct current transmission according to claim 2, characterized in that: Step S1 uses the voltage fluctuation at the common connection point as the objective function to measure the stability of the system voltage. The calculation formula is: Where: U pcc is the actual voltage at the node common connection point, is the expected voltage at the common connection point.

4. The method for dynamic optimization of reactive power and voltage taking into account wind and thermal power generation via direct current transmission according to claim 3, characterized in that: Step S2 measures the severity of the transient overvoltage at the node under transient conditions by using the transient voltage over-limit index, which is expressed as: Where: U(t) is the instantaneous voltage value of the sending end busbar under transient state, U s (t) is the voltage stability limit of the sending end busbar, t k is the time sequence of the integral discrete points arranged in the direction of the integral interval, Δt k N is the width corresponding to each discrete point. K To calculate the total number of discrete points within the time period; The calculation formula for transient overvoltage suppression index is: Where: K * is the initial value of the operating parameter, and ΔK is the change in the adjusted operating parameter.

5. The method for dynamic optimization of reactive power and voltage taking into account wind and thermal power generation via direct current transmission according to claim 4, characterized in that: The structure of the parameter dynamic optimization control system in step S2 is: The parameter dynamic optimization control system structure in step S2 is: Deep fuzzy neural controllers A and B are used to update and optimize the PI controller parameters of the outer and inner loops of the STATCOM in real time. The input of deep fuzzy neural controller A is the error between the rated value and the measured value of the line voltage at the grid connection point, the error change rate and the wind speed, and the output is the optimization instruction ΔK of the PI controller. pA , ΔK iA The input of deep fuzzy neural controller B is the error between the reference value and the measured value of the q-axis current output by the outer loop, the error change rate and the wind speed, and the output is the PI controller optimization instruction ΔK pB , ΔK iB , the proportional gain K of the PI controller p and integral gain K i According to formula (6), it is updated at each sampling moment: In formula (6), and They represent the proportional and integral empirical parameters of the STATCOM outer and inner loop PI controllers, ΔK p and ΔK i They respectively represent the real-time generated dynamic optimization instructions for the proportional and integral coefficients.

6. The method for dynamic optimization of reactive power and voltage taking into account wind and thermal power generation via direct current transmission according to claim 5, characterized in that: Step S3: Deep fuzzy neural network uses error back propagation algorithm to update parameters. First, define S h is the hidden layer of the network, the input x=[x1,x2,…,x d ,…,x D ], the network parameters involved are: Where: θ h is the parameter matrix of the fuzzy membership of the input condition variables in the fuzzy rules; is the input to the hidden layer S h The parameter of the membership degree of the d-th conditional clause of the f-th rule on h is the coefficient matrix of the conclusion part of the fuzzy rule; is the parameter of the membership degree of the dth conditional clause of the fth rule in the conclusion of the fuzzy rule; μ h is the hidden layer S h The membership matrix corresponding to the node, is the hidden layer node S h The membership value of the d-th conditional clause of the corresponding r-th rule, where r = 1, 2, ..., R, D represents the number of inputs to the hidden layer, and F represents the number of inputs to the hidden layer S h The number of fuzzy rules, R represents the number of hidden layer nodes S h The number of fuzzy rules.

7. The method for dynamic optimization of reactive power and voltage taking into account wind and thermal power generation via direct current transmission according to claim 6, characterized in that: Step S3 calculates the output values ​​of all hidden layer nodes in the network. First, it is necessary to determine the membership weights of the rule conditions. Since the condition part in the fuzzy rule is composed of multiple conditional clauses, it is necessary to use the t operator to merge the membership values ​​of each clause input variable, which is specifically expressed as: Where: is the hidden layer node S h The weight of the corresponding r-th rule, D represents the number of inputs; The output of the nodes in the network is calculated based on the conclusion part of the fuzzy rules. The r-dimensional vector composed of the output of all rules is: Where: is the hidden layer node S h The output of the corresponding rth rule, v h is the hidden layer node S h The corresponding output vector matrix of all fuzzy rules, is the hidden layer node S h The coefficient of the d-th conditional clause of the r-th rule corresponding to the conclusion of the fuzzy rule; Network hidden layer node S h The overall output can be expressed as the weighted average sum of the weights and node outputs:

8. The method for dynamic optimization of reactive power and voltage taking into account wind and thermal power generation via direct current transmission according to claim 7, characterized in that: After obtaining the outputs of all nodes in the network, step S3 defines the loss function of the entire network as the objective function. The minimum loss function is the goal of the entire network parameter learning, which is defined as: Where: e is the output error, y is the output value of the network, y d is the target value, J is the loss function, e (n) is the error of the nth group of samples, and N is the total number of samples.

9. The method for dynamic optimization of reactive power and voltage taking into account wind and thermal power generation via direct current transmission according to claim 8, characterized in that: Step S3 calculates the loss according to the overall loss function in the network, and then propagates the error of the output layer node back to the entire network, and calculates the gradient of the loss function on each network parameter, and finally updates all parameters in the network by gradient descent method; The calculation method of the gradient of the loss function on the output layer node is: In formula (15)(16), is the coefficient of the hth conditional clause of the rth rule in the conclusion part of the fuzzy rule corresponding to the output layer node O, is the output corresponding to the rth rule of the output layer node O, is the weight of the rth rule of the output layer node O.

10. A method for dynamic optimization of reactive power and voltage taking into account wind and thermal power generation via direct current transmission according to claim 9, characterized in that: The calculation of the gradient of the hidden layer parameters in step S3 is similar to the calculation of the gradient of the output layer y. After the gradients corresponding to all parameters in the network are calculated, the parameters can be updated synchronously, which is specifically expressed as: Where: ψ is all network parameters, where ψ new and ψ old Represent the network parameters before and after the update respectively; α is the rate of parameter learning.

Images