A data-driven converter station steady-state reactive power coordination strategy optimization method and device
Through the data-driven method, the SVR prediction model and particle swarm algorithm are used to optimize the reactive power output of the camera and AC filter, which solves the problem of insufficient dynamic reactive power support capability in the reactive power coordination strategy of the converter station and improves the grid voltage Stability and operational economics.
Patent Information
- Application Number
- CN202210474397.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-04-29
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2042-04-29
AI Technical Summary
The existing technology has not fully considered the impact of the coordination of different reactive power compensation devices of the converter station on the voltage stability of the converter bus, resulting in the economic operation of the system still needs to be improved, especially when the dynamic reactive power support capacity is insufficient during large DC transmission, which can easily cause the power network. Low voltage or voltage crash.
Based on the data-driven method, the transient stability margin prediction model of the converter bus voltage is established by supporting vector regression (SVR), and combined with the particle swarm algorithm to optimize the steady-state reactive power output of the camera and AC filter to maximize the converter bus voltage. Transient stability margin and reduce system network loss and reduce the number of AC filter input groups.
Effectively improve the transient stability of the converter bus voltage, reduce the system network loss, reduce the number of input groups of AC filters, improve the operating economy and safety of the power grid, avoid long-term transient stable calculations and The voltage stability is complex and difficult to analyze mathematically.
Smart Images

Figure CN114759590B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to a reactive power control optimization method, and in particular to a data-driven converter station steady-state reactive power coordination control strategy optimization method. Background Art
[0002] With the continuous development and construction of long-distance, large-capacity UHV AC / DC transmission projects in my country, large-scale DC feed-in has made the contradiction of "strong DC and weak AC" in the power grid prominent. In the dynamic process of AC system failure causing DC commutation failure, a single DC can absorb 8 million kvar reactive power or even more, which can easily cause continuous low voltage or even voltage collapse and DC lockout in the system, seriously threatening the safe operation of the power grid. Therefore, in order to improve the voltage stability of the UHV DC power grid at the transmitting and receiving ends, it is necessary to match the corresponding dynamic reactive support when DC large-scale active power is transmitted, that is, "large DC transmission and strong reactive support". As a synchronous rotating device, the phase regulator can not only provide short-circuit capacity for the system, but also has the characteristics of strong excitation reactive output. In recent years, it has been applied to DC converter stations to improve the voltage stability of the commutation busbar. In order to further improve the stability of the DC power grid at the transmitting and receiving ends, many scholars and engineering and technical personnel have carried out relevant technical research and application.
[0003] At present, some studies focus on the optimal configuration and control operation of DC systems and phase regulators to improve the system's dynamic reactive power support capacity and reduce the risk of DC commutation failure. Some studies point out that the hierarchical access method will help improve the voltage support capacity of multi-input DC systems and guide the reasonable distribution of power flow; some studies comprehensively evaluate the impact of adding phase regulators on multi-DC feed-in receiving-end power grids, providing effective reference and basis for the capacity and parameter configuration of synchronous phase regulators; some studies optimize phase regulator parameters based on frequency domain sensitivity analysis to significantly improve their dynamic reactive power characteristics; some studies design and propose a phase regulator emergency control architecture and method, which can be used for emergency excitation and continuous increase in reactive power after commutation failure. However, these studies have failed to consider the impact of different reactive compensation devices in the converter station on the voltage stability of the commutation bus, so the economic operation of the system still needs to be improved. For this reason, some studies have proposed that the synchronous phase regulator can compensate for reactive power by replacing the AC filter with adjustable residual capacity to make it emit a certain amount of reactive power in a steady state, so as to improve the voltage safety level of the power grid with a smaller dispatching cost, but the reactive power output of the phase regulator is limited by the voltage level of the commutation bus before the accident, and the advantages of the new generation of phase regulators have not been fully utilized.
[0004] Aiming at the shortcomings of the current research on reactive power coordination strategies and methods of converter stations, this application compares the differences in transient reactive power characteristics between AC filters and synchronous phase regulators and the influence of their coordination on the transient stability of the commutation bus voltage. Based on the number of AC filter groups, the steady-state output of the phase regulator, the reactive power exchange amount, the steady-state voltage of the commutation bus, and the transient stability margin index η of the commutation bus voltage based on a multi-binary table, a novel reactive power coordination strategy for converter stations is proposed. VThe data is driven by the support vector regression (SVR) method to establish an effective commutation bus η V Prediction model is used to solve the complex mathematical analysis difficulties of long-term transient stability calculation of large power grid and bus voltage stability. V The prediction model is based on the optimization objectives of maximizing the transient stability margin of the commutation bus voltage and reducing the system network loss. The steady-state reactive output optimization model of the condenser and AC filter of the commutation station is established. Finally, MATLAB is used to call PSASP for power flow calculation, and the particle swarm algorithm is combined to quickly solve it. The invented method can make the synchronous condenser cooperate with the AC filter to achieve the best reactive output, reduce the number of AC filter input groups, and effectively improve the transient stability of the commutation bus voltage. Finally, the reactive coordination strategy optimization method studied was simulated and verified. Summary of the invention
[0005] The above technical problems of the present invention are mainly solved by the following technical solutions:
[0006] A data-driven converter station steady-state reactive power coordination strategy optimization method, characterized in that:
[0007] Based on the reactive output of the condenser phase in the converter station, the number of AC filter groups, the reactive exchange between the converter station and the DC feed-in grid, and the steady-state voltage of the converter bus, a multi-dimensional input feature vector of the regression model for predicting the transient stability margin of the converter bus voltage is constructed, and the support vector regression method is used to establish the prediction model of the transient stability margin of the converter bus voltage.
[0008] Taking maximizing the transient stability margin of the commutation bus voltage and reducing the system network loss as the optimization goals, the coordinated optimization objective function of the reactive output of the phase regulator and AC filter of the converter station is established to take into account both the improvement of the transient stability of the commutation bus voltage and the economic efficiency of the system operation. In addition, additional constraints such as the number of AC filter groups in the converter station, the reactive exchange amount of the commutation bus, the reactive output of the phase regulator, and the node voltage are added to establish an optimization model for the steady-state reactive output of the converter station.
[0009] The steady-state reactive power optimization model of the converter station is solved based on power flow calculation and particle swarm algorithm, which specifically includes: taking the phase converter output and the number of filter groups as the optimal particles, initializing the particle swarm parameters and performing power flow analysis and calculation of the DC feed-into the power grid. For converged power flows and the power flow calculation results satisfying the additional constraints of the optimization model, a prediction model is constructed based on the power flow calculation results, mostly as input feature quantities, and the commutation bus voltage temporary stability margin prediction model and the objective function are used to update the individual particle optimal and the population optimal. Otherwise, the particle speed and position are directly updated and the cycle is repeated until the number of iterations reaches the preset number and the optimal solution is output.
[0010] In the above-mentioned data-driven converter station steady-state reactive power coordination strategy optimization method, the minimum value of the converter bus voltage transient stability margin index under multiple expected severe faults (f = 1, ..., n) of the DC feed-in power grid is selected. As the transient stability margin of the commutation bus voltage, a quantitative evaluation index of the voltage transient stability margin based on multi-binary table and weighted integration is proposed The definition is as follows:
[0011]
[0012] Where: V N is the bus reference voltage, V(t) is the bus transient voltage under fault disturbance; t p and t′ p are the voltages below and above the pth voltage critical reference value V during the drop and recovery process of V(t), respectively. cr,p The moment of K p is(t p ,t p+1 )∪(t′ p+1 ,t′ p ) interval integral weight coefficient; when V(t) is less than V cr,1 ,…,V cr,p ,…,V cr,q When the voltage drop and recovery process is located at (t1, t2)∪(t′2, t′1),…,(t p ,t p+1 )∪(t′ p+1 ,t′ p ),…,(t q ,t′ q ) The integral weight coefficients of the time interval are K1,…,K p ,…,K q ; The weight coefficient of each interval can be based on the voltage transient response, in order to keep the voltage critical transient stability, that is, η V =0, determined by solving formula (1) step by step;
[0013]
[0014] Where, T cr,q For voltages lower than V cr,q According to the transient voltage stability criterion determined by the Technical Specification for Power System Safety and Stability Calculation issued by the National Grid Operation and Control Standardization Technical Committee, the multi-binary table critical voltage and integral weight coefficient of the quantified bus voltage temporary stability margin shown in Table 1 can be determined.
[0015] Table 1 Multi-binary table critical voltage and integral weight coefficient (V N =1.0pu)
[0016]
[0017] In the above-mentioned data-driven converter station steady-state reactive power coordination strategy optimization method, the following steps are adopted to establish a support vector regression model for predicting the transient stability margin of the converter bus voltage based on the multi-dimensional input feature vector based on the transient stability calculation of the power grid expected severe faults:
[0018] Step 1: Adjust the number of phase regulators and AC filters in the converter station and calculate the power flow. At the same time, perform transient stability calculations on the converged power flows after reactive power regulation according to the expected severe fault set.
[0019] Step 2: construct a multidimensional input feature vector according to the calculation results of each convergent power flow, and determine the transient stability margin of the commutation bus voltage of each convergent power flow based on the transient stability calculation results, so as to construct machine learning sample data of a support vector regression model for predicting the transient stability margin of the commutation bus voltage based on the multidimensional input feature vector;
[0020] Step 3: The Z-score normalization method shown in formula (2) is used to normalize the components of the multidimensional input feature vector and the transient stability margin of the commutation bus voltage in the data sample set. After data preprocessing, the samples are divided into a training set and a test set using the classic 7:3 ratio;
[0021]
[0022] In the formula, x m,i and are the i-th sample X i The mth component and its normalized value, sample X i The vector is composed of the reactive power output of the phase regulator, the number of AC filter groups, the reactive exchange between the converter station and the DC feed-in grid, the steady-state voltage of the commutation bus, and the transient stability margin of the commutation bus voltage. and are the mean and standard deviation of the mth component respectively;
[0023] Step 4: Set the training set samples X i Substitute the vector into the optimized mathematical model of the commutation bus voltage transient stability margin prediction model based on support vector machine regression shown in formula (III), use k-fold cross validation to perform grid global parameter optimization on the model parameters c and g, and finally establish the commutation bus voltage transient stability margin prediction model based on the training set data, and determine the final prediction model after evaluating the learning effect of the model based on the test set data;
[0024]
[0025] In the formula, α i and represents a Lagrangian operator greater than or equal to zero, X i and X j are the i-th and j-th samples of the training set, l is the number of samples in the training set, ε is the pipeline threshold that describes the insensitive deviation range between the model predicted output and the actual value of the transient stability margin of the commutation bus voltage, c is the regularization constant, and the larger the c, the greater the penalty for the sample points that exceed the ε pipeline. K(X i ,X j ) is a Gaussian radial kernel function, which is used to calculate X in the original sample space. i and X j The inner product in the high-dimensional feature space enables regression learning to be performed implicitly in the feature space, and g is the kernel parameter.
[0026] In the above-mentioned data-driven converter station steady-state reactive power coordination strategy optimization method, it is considered that the DC transmission power and the reactive voltage regulation equipment fed into the grid remain unchanged in the initial state during the steady-state operation of the power grid, and at the same time, the constraints such as the reactive exchange of the converter station, the minimum number of filter groups of the AC filter and the reactive capacity of the phase regulator are considered. By optimizing the steady-state reactive output of the phase regulator and the number of AC filter groups, the optimization model shown in formula (IV) is established with the goal of maximizing the transient stability margin of the converter bus voltage and reducing the system network loss.
[0027]
[0028] Where P loss is the power grid loss; λ1 and λ2 are the optimization target weight coefficients, whose values are assigned according to the optimization target bias and the order of magnitude of the optimization target value change; Q scN N is the rated capacity of a single phase regulator; min The minimum input group of AC filter; Q cN is the rated capacity of a single filter group; V max 、V min They are the upper and lower limits of the commutation bus voltage respectively.
[0029] In the above-mentioned data-driven converter station steady-state reactive power coordination strategy optimization method, the solution process of the reactive power coordination strategy optimization model includes:
[0030] S1. Initialize the particle swarm parameters, including setting the weight parameters c1 and c2 that reflect the particle's own experience and social experience, setting the particle's inertia weight parameter ω∈[0,1] to maintain the existing optimization speed, and setting the dimension of the optimization target search space equal to the number of reactive power optimization variables of the converter station. Set the optimization space S according to the number of AC filter groups and the rated capacity of the phase shifter. n ;vector is the optimal particle i, where the component represents the reactive output of the phase regulator of the optimal particle i, and the component Represents the number of AC filter groups of the optimal particle i, defines the number of particles as r, and in the optimal space S n The initial components of each particle are randomly generated in At the same time, the initial optimization speed of each particle is randomly generated The optimal fitness value of each particle to the objective function The optimal fitness value of the particle swarm is Fit p =∞; the maximum number of iterations of the optimal search is k max , and set the initial value of the iteration number k to 1;
[0031] S2. Update the output reactive power of the condenser and AC filter of the converter station based on each optimal particle and calculate the power flow. For each converged power flow, construct a multi-dimensional input feature vector based on the power flow result and predict the temporary stability margin of the commutation bus voltage. At the same time, calculate the objective function of the optimization model shown in formula (IV) and analyze the fitness value of each particle in the current kth iteration. And judge it, if but And make Otherwise, keep the optimal fitness value of each particle With the best position unchanged, and at the same time, the optimal fitness value of each particle is maintained for each non-convergent trend With the best position unchanged, after all particles have been optimally updated, if Fit p Updated to And update The corresponding particle is the global optimal particle Otherwise keep Fit p and constant;
[0032] S3, use the particle velocity and displacement update formula to calculate the particle's spatial position at the k+1th iteration. If k+1 <k max , then k=k+1, return to S2 to perform the next iterative particle optimization calculation until the number of iterations reaches k max , output the optimal solution.
[0033] A data-driven converter station steady-state reactive power coordination strategy optimization device, characterized by comprising:
[0034] Module 1: Based on the reactive output of the condenser phase in the converter station, the number of AC filter groups, the reactive exchange between the converter station and the DC feed-in grid, and the steady-state voltage of the converter bus, a multi-dimensional input feature vector of the regression model for predicting the transient stability margin of the converter bus voltage is constructed, and the support vector regression method is used to establish the prediction model of the transient stability margin of the converter bus voltage;
[0035] The second module: Taking the maximization of the transient stability margin of the commutation bus voltage and the reduction of the system network loss as the optimization goals, the coordinated optimization objective function of the reactive output of the phase regulator and AC filter of the converter station is established to take into account both the improvement of the transient stability of the commutation bus voltage and the economic efficiency of the system operation, and additional constraints such as the number of AC filter groups in the converter station, the reactive exchange amount of the commutation bus, the reactive output of the phase regulator, and the node voltage are added to establish an optimization model for the steady-state reactive output of the converter station;
[0036] The third module: Solve the steady-state reactive power optimization model of the converter station based on power flow calculation and particle swarm algorithm, specifically including: taking the output of the condenser and the number of filter groups as the optimal particles, initializing the particle swarm parameters and performing power flow analysis and calculation of the DC feed-into the power grid. For converged power flows and the power flow calculation results satisfying the additional constraints of the optimization model, a prediction model is constructed based on the power flow calculation results, mostly as input feature quantities, and the commutation bus voltage temporary stability margin prediction model and the objective function are used to update the individual particle optimality and the population optimality. Otherwise, skip directly updating the particle speed and position, and repeat this cycle until the number of iterations reaches the preset number and the optimal solution is output.
[0037] Therefore, the present invention is characterized in that the commutation bus voltage transient stability margin η is established based on data drive. V The prediction model was established, and the steady-state reactive power output optimization model of the converter station was established with the optimization objectives of maximizing the transient stability margin of the commutation bus voltage and reducing the system network loss. Finally, MATLAB was used to call PSASP for power flow calculation, and the particle swarm algorithm was combined for rapid solution to establish the optimization method of the steady-state reactive power coordination control strategy of the converter station. The objective function also takes into account the economy of the system and the constraints such as the reactive exchange of the commutation bus. It can give full play to the dynamic reactive power support capacity of the phase regulator, reduce the investment in AC filters, and effectively improve the transient stability of the commutation bus voltage. By establishing the commutation bus η V The prediction model avoids the difficulty of complex mathematical analysis of long-term transient stability calculation of large power grids and commutation bus voltage stability, and provides the possibility and rapidity for solving the reactive power output optimization model. BRIEF DESCRIPTION OF THE DRAWINGS
[0038] Figure 1 is the temporary stability margin of the commutation bus voltage η V Prediction model building process;
[0039] Figure 2Solve the calculation process for reactive power coordination optimization model;
[0040] Figure 3 The 500kV grid structure diagram of Jiangxi Power Grid and the structure diagram of Nanchang Converter Station;
[0041] Figure 4a is the commutation bus η V Prediction model learning effect evaluation results (error pie chart);
[0042] Figure 4b is the commutation bus η V Evaluation results of the learning effect of the prediction model (error data graph);
[0043] Figure 5 Comparison of transient curves of Sanyong fault on Yunfeng-Chongren I line under two reactive power control methods;
[0044] Figure 6 Comparison of transient curves of Sanyong fault on Nanchang-Jinxian line under two reactive power control methods DETAILED DESCRIPTION
[0045] The technical solution of the present invention is further specifically described below through embodiments and in conjunction with the accompanying drawings.
[0046] 1. Evaluation of Transient Stability Margin of Commutation Bus Voltage
[0047] Quantitative evaluation index of voltage transient stability margin based on multi-binary table and weighted integral Taking into account the impact of voltage sag duration, the severity of AC system faults is described more finely. The indicator is defined as follows:
[0048]
[0049] Where V N is the commutation bus reference voltage, V f (t) is the real-time transient voltage of the busbar after the fault f is removed; t p and t' p V f (t) During the drop and recovery process, the voltage is lower than and higher than the pth voltage critical reference value V cr,p The moment when V f (t) are located at V cr,q Below, ..., V cr,p+1 ~V cr,p ,…,V cr,2 ~V cr,1 , that is, during the voltage drop and recovery process, they are respectively located at (t q ,t' q ),…,(t p ,t p+1 )∪(t'p+1 ,t' p ), …, (t1, t2) ∪ (t'2, t'1), assign different voltage interval integral weight coefficients K q , …, K p …, K1.
[0050] The weight coefficient in formula 1 can be calculated based on the voltage dynamic process, and the critical stable state of the system satisfies The step-by-step solution is as follows:
[0051]
[0052] Where: T cr,q For voltages lower than V cr,q The maximum acceptable duration.
[0053] At the same time, considering that the research goal of this application is to maximize the voltage stability of the converter bus of the converter station to resist the impact of severe system fault disturbances and reduce the risk of voltage instability and DC commutation failure, the transient stability calculation of this application selects the minimum value of the transient voltage stability margin index of multiple severe faults (f=1,…,n) as the transient voltage drop stability margin of the converter bus.
[0054]
[0055] According to the relevant voltage standards and the operation experience of the power grid, the criterion of the multi-binary table of commutation bus voltage and the corresponding weight coefficients are set as shown in Table 1.
[0056] Table 1 Criteria and solution weight coefficients for multiple binary tables
[0057]
[0058] 2. Transient stability margin of commutation bus voltage η V Regression prediction model establishment
[0059] According to the reactive power exchange principle of the commutation bus at the receiving end converter station, the key steady-state operating factors affecting the transient stability of the commutation bus voltage are the total reactive power output of the AC filter, the reactive power output of the phase regulator, and the reactive power provided by the AC system (i.e., the reactive power exchange amount of the converter station). The total reactive power output of the AC filter is related to the number of filter groups put into use and the steady-state voltage of the commutation bus. Therefore, this application considers the reactive power output Q of the phase regulator to be sc0 , the number of AC filter groups N, the reactive exchange amount of the converter station Q exc , the commutation bus steady-state voltage U0 is used as the characteristic input quantity, and the commutation bus steady-state voltage stability margin η V As the prediction output, the support vector regression method is used to establish η V Predictive models.
[0060] Assume that the given nonlinear inseparable training data set is {(x i ,y i ),i=1,2,…l)}, where x i is the feature input vector, y i is the corresponding target value, and l is the number of samples. The support vector regression learning strategy is to find the best separation hyperplane (i.e., regression estimation function) to minimize the comprehensive distance from all sample target values to the plane. Its dual optimization problem is as follows:
[0061]
[0062] Where: l is the number of samples, c is the penalty parameter, α=(α1,α2,...,α l ) T is the Lagrange multiplier vector, α * =(α1 * ,α2 * ,...,α l * ) T is the solution to the dual optimization problem, K(x i ,x j )=φ(x i )φ(x j ) is called the kernel function,.
[0063] This application uses the commonly used Gaussian radial kernel function, and the calculation formula is as follows:
[0064]
[0065] Where: g is the kernel parameter.
[0066] Therefore, the commutation bus voltage temporary stability margin η V Process such as Figure 1 shown.
[0067] When batch simulation calculation is performed in advance to obtain the steady-state operation information of the converter station and the temporary stability margin of the converter bus voltage η V After the sample data is constructed, data preprocessing is required, including removing outliers and normalizing the data. In order to avoid the problem of difficult convergence caused by the difference in numerical dimensions and sample distribution between features, the Z-score standardization is used to normalize the steady-state operation data of the converter station and the temporary stability margin of the converter bus voltage collected in the sample generation stage:
[0068]
[0069] In the formula, is a certain-dimensional data vector x of the i-th sample data i The average value of is the vector x iThe standard deviation of .
[0070] After data preprocessing, the samples are randomly divided into training set and test set using the classic 7:3 ratio. The key hyperparameters that affect the accuracy of the prediction model are c and g parameters. In order to improve the performance of the model and avoid the model from falling into overfitting or underfitting, this application uses k-fold cross validation to perform grid global parameter optimization on the model, and finally establishes the commutation bus η based on the training set data. V Predict the model and evaluate the learning effect of the model based on the test set data.
[0071] 3. Establishment of steady-state reactive power optimization model for converter stations
[0072] This application takes the DC transmission power remaining unchanged and the discrete voltage regulation equipment maintaining the initial state unchanged as the basic assumptions, takes the maximization of the transient stability margin of the commutation bus voltage and the reduction of the system network loss as the optimization goals, and establishes the reactive power optimization objective function of the converter station transient voltage preventive control and system operation economy according to Formula 7.
[0073] min F(Q sc0 ,N)=α1P loss -α2η V Formula 7
[0074] In the formula, Q sc0 is the steady-state reactive power output of the phase regulator, N is the number of filter groups, P loss is the system network loss, η V is the transient stability margin of the commutation bus voltage, α1 and α2 are the optimization target weight coefficients, and their sizes are assigned according to the optimization target bias and the order of magnitude of the optimization target value change.
[0075] The constraints for the optimization of the reactive power coordination strategy of the converter station are established: in addition to the AC filter in the converter station meeting the minimum number of filter groups, the reactive exchange volume of the converter bus should be limited to a certain range to ensure that the steady-state voltage of the converter bus operates within the normal range. At the same time, the reactive output capacity limit of the phase-shifting machine and the node voltage constraints are added to construct the optimization constraints for the reactive power coordination strategy of the converter station shown in Formula 8.
[0076]
[0077] Where: Q scN is the rated capacity of a single phase regulator, N min is the minimum number of filter groups, Q c_sum is the total reactive power compensation of all filters, Q cN is the rated capacity of a single filter group, Q dc Reactive power consumed by the converter station, V max 、V min are the upper and lower limits of the node voltage respectively.
[0078] 4.MATLAB calls PSASP power flow calculation and combines the model solution process of particle swarm algorithm
[0079] This application uses MATLAB to call PSASP to calculate the power flow of the large power grid to obtain steady-state operation information, and then uses the above-established commutation bus voltage transient stability margin prediction model to replace the time domain transient stability simulation calculation, and combines the particle swarm optimization algorithm to solve the above-established commutation station steady-state reactive power output optimization model.
[0080] The particle swarm optimization algorithm simulates the predation process of flying birds and uses memory and feedback mechanisms to achieve efficient optimization search. Its basic idea is that each particle moves in the solution space, and in each optimization process, each particle records the optimal point x that it has searched. ps , and record the global optimal point x searched by all particles g At the same time, the particle updates its speed and position according to its own optimal point and the global optimal point, and the cycle continues until it converges to the global optimal point.
[0081] In order to avoid premature convergence and insufficient solution accuracy of the particle swarm optimization algorithm, this application adopts the particle speed and position update rules in the particle swarm optimization algorithm as follows:
[0082]
[0083] In the formula, x s (k), x s (k+1) are the individual positions of particles before and after iterative update; v s (k), v s (k+1) are the individual particle velocities before and after the update; x ps (k), x g (k) are the optimal position of the individual particle and the optimal position of the population before updating; c1 and c2 are learning factors; r1 and r2 are random numbers between 0 and 1; ω s is the inertia weight factor, and the iteration adopts a linear decreasing strategy, ω int ,ω end are the initial and final values of the inertia weight factor iteration respectively; k is the current iteration number, k max is the maximum number of iterations.
[0084] The flowchart of the reactive power coordination strategy optimization model is as follows: Figure 2 As shown, the specific steps are as follows:
[0085] (1) Set model parameters.
[0086] (2) Initialization, setting acceleration constants c1, c2, and maximum evolutionary number k max, set the current evolutionary generation to k = 1; randomly generate m particles x1, x2, ..., x m , forming the initial population x(0); randomly generating the initial velocities v1, v2, ..., v m , forming the velocity displacement matrix v(0).
[0087] (3) Set the condenser steady-state output Q sc0 The number of AC filter groups N is set as particles, and the particles assigned by the particle swarm algorithm are substituted into the programming package of MATLAB to call the PSASP power flow subroutine to perform power flow calculations to determine whether the power flow converges and satisfies the additional constraints of the objective function. If satisfied, execute the next step; if not satisfied, replace the current solution with the nearest feasible solution and execute the next step.
[0088] (4) Calculate the objective function, compare the current fitness and the best fitness in memory, if the current fitness is better than the best fitness in memory, replace it and record the current position as the best position of the particle. Compare the best position of the particle with the best position of the population, if the best position of the particle is better than the best position of the population, replace it, otherwise, keep it unchanged.
[0089] (5) Particle velocity and position update.
[0090] (6) Return to (3) and repeat the calculation until the maximum number of iterations is met, and then output the optimal solution.
[0091] 5. Simulation Example
[0092] This application adopts Figure 3The feasibility of the proposed reactive power coordination optimization method is demonstrated for the Nanchang converter station at the receiving end of the Yazhong-Jiangxi UHV DC shown in the figure. The planned rated capacity of the Yazhong DC is 8000MW, and it is connected to the Jiangxi power grid in layers. The high end of the converter station is connected to the Nanchang transformer and Fuzhou transformer through 4 500kV outgoing lines, and the low end is planned to be connected to the medium voltage side of the Nanchang UHV station through 3 interconnection lines. The high end of the converter station is equipped with 12 groups of filter capacitors, with a single group rated capacity of 240Mvar; the low end of the converter station is equipped with 10 groups of filter capacitors, with a single group rated capacity of 270Mvar. The high and low ends of the converter station are each equipped with a synchronous phase regulator with a rated capacity of 300Mvar. The example comes from the transitional power grid data for the 2021-2022 period uniformly issued by the National Electric Power Dispatching and Communication Center. In this data packet, since the actual Huazhong UHV AC ring network has not yet been put into operation, only the high end of the Nanchang converter station at the receiving end of the Yazhong DC is put into operation. The simulation condition of this application is to set the contact switch between the high and low ends of the closed converter station to make the high and low end synchronous phase regulators run in parallel (with the same reactive output), and the low end AC filter is fixed to 6 groups, and only the steady-state reactive coordination control strategy of the phase regulator and the high end AC filter is studied to improve the stability of Jiangxi power grid. The specific setting of Ya'an-Zhong DC transmission power is 2750MW, and the minimum number of high-end filter input groups is 1 group. The system base capacity is 100MV·A, and the 500kV system base voltage is 525kV. The temporary stability simulation duration is 10s. The selected Jiangxi power grid expected serious fault set is shown in Table 2. All faults are three-phase permanent phase faults. The fault occurrence time is t=1.0s, and the removal time is t=1.1s.
[0093] Table 2 Severe faults in Jiangxi power grid selected for temporary stability calculation
[0094]
[0095] According to this paper, MATLAB programming is used to modify the reactive output parameters of the phase-converter and the number of AC filter input groups in the PSASP simulation input file, and PSASP is called to perform batch calculations of power flow and temporary stability to obtain the steady-state operation information of the Nanchang converter station and the commutation bus η V The raw data sample of the indicator. Figure 1 The process shown in the figure establishes the commutation bus η V Prediction model, and evaluate the learning effect of the built prediction model. The results are as follows Figure 4a and Figure 4b shown.
[0096] Depend on Figure 4a It can be seen that the voltage transient stability margin η of the sample test set is V The predicted value is close to the true value, and the number of samples with a relative error within 5% accounts for 95%, which has high accuracy. Figure 4b It can be seen that the samples with large relative errors between the predicted value and the true value are mainly distributed in the commutation bus ηV For points where the true value is significantly smaller, the predicted value is still very small, that is, the commutation bus voltage is seriously unstable and collapsed, so it will not affect the subsequent optimization process with the voltage transient stability margin as large as possible as the optimization goal.
[0097] according to Figure 2 The steady-state reactive power coordinated control process of the converter station shown in the figure sets the objective function weight coefficients α1=1, α2=10, PSO learning factor c1=c2=2, number of particles m=10, and the maximum number of iterations is k max =50, and the optimal solution of the reactive power output model of the converter station is the steady-state output Q of the phase regulator. sc0 =2.944pu, number of AC filter groups N = 1.
[0098] The conventional control strategy of the converter station adopts the automatic control input of the AC filter, the steady-state reactive output of the phase regulator is zero, and the voltage is maintained near the rated value. The voltage stability of the Jiangxi power grid under the proposed strategy and the conventional strategy is compared. The stable operation status of the Nanchang converter station under the two methods is shown in Table 3. It can be seen from Table 3 that compared with conventional control, coordinated control can reduce the number of AC filter input groups, maintain the steady-state voltage within the normal range, and the objective function value is better.
[0099] Table 3 Nanchang converter station operation status
[0100]
[0101] Now, the 500kV Yunfeng-Chongren I line and Nanchang-Jinxian line of Jiangxi Power Grid are set with N-1 three-phase permanent AC fault respectively. The transient stability curves of high-end converter bus voltage and high-end converter valve shut-off angle of Nanchang converter station are as follows: Figure 5 , Figure 6 shown.
[0102] Depend on Figure 5 It can be seen that after the fault of Yunfeng-Chongren I line, under both control strategies, the high-end converter valve only had one commutation failure, and the high-end converter bus voltage could be restored to a stable initial value. However, the proposed strategy provided more reactive power during the transient period of the down-regulation phase, which made the voltage recover to a stable state earlier.
[0103] Depend on Figure 6It can be seen that after the fault of the Nanchang-Jinxian line, under the conventional control strategy, the phase regulator does not provide reactive power in the initial steady state, and the AC filter mainly performs reactive power compensation. Therefore, when the voltage drops severely, the system's dynamic reactive power support capacity is insufficient and cannot support voltage recovery, and the converter valve continues to fail to commutate. Under the control of the proposed strategy, the phase regulator fully exerts its dynamic reactive power support capacity, the voltage can be restored to normal operation, and the converter valve only fails to commutate once. It can be seen that the proposed strategy not only reduces the number of steady-state AC filter input groups, but also can give full play to the reactive power support capacity of the phase regulator after the system fault disturbance, so that the commutation bus voltage recovery speed and stability are better.
[0104] It can be seen that compared with the conventional method, the proposed converter station reactive power coordinated control method can reduce the number of AC filter groups, give full play to the reactive power support advantages of phase regulators, improve the voltage transient stability of the commutation bus, and make the power system operation safer and more economical.
[0105] The specific embodiments described herein are merely examples of the spirit of the present invention. Those skilled in the art may make various modifications or additions to the specific embodiments described or replace them in similar ways, but they will not deviate from the spirit of the present invention or exceed the scope defined by the appended claims.
Claims
1. A data-driven converter station steady-state reactive power coordination strategy optimization method, characterized in that: Based on the reactive output of the condenser phase in the converter station, the number of AC filter groups, the reactive exchange between the converter station and the DC feed-in grid, and the steady-state voltage of the converter bus, a multi-dimensional input feature vector of the regression model for predicting the transient stability margin of the converter bus voltage is constructed, and the support vector regression method is used to establish the prediction model of the transient stability margin of the converter bus voltage. Taking maximizing the transient stability margin of the commutation bus voltage and reducing the system network loss as the optimization goals, the coordinated optimization objective function of the reactive output of the phase regulator and AC filter of the converter station is established to take into account both the improvement of the transient stability of the commutation bus voltage and the economic efficiency of the system operation. In addition, additional constraints such as the number of AC filter groups in the converter station, the reactive exchange amount of the commutation bus, the reactive output of the phase regulator, and the node voltage are added to establish an optimization model for the steady-state reactive output of the converter station. Solving the steady-state reactive power optimization model of the converter station based on power flow calculation and particle swarm algorithm, specifically including: taking the output of the condenser and the number of filter groups as the optimal particles, performing power flow analysis and calculation of the DC feed-into the power grid after initializing the particle swarm parameters, and constructing the multi-dimensional input characteristic quantity of the prediction model according to the power flow calculation results for converged power flow and the power flow calculation results satisfying the additional constraints of the optimization model, and using the commutation bus voltage temporary stability margin prediction model and the objective function to update the individual optimal and population optimal particles, otherwise skipping and directly updating the particle speed and position, and repeating this cycle until the number of iterations reaches the preset number of times and the optimal solution is output; Considering that the DC transmission power and the reactive voltage regulation equipment fed into the grid remain unchanged in the initial state during the steady-state operation of the power grid, and considering the constraints such as the reactive exchange of the converter station, the minimum number of filter groups of the AC filter and the reactive capacity of the phase regulator, the optimization model shown in formula (IV) is established by optimizing the steady-state reactive output of the phase regulator and the number of AC filter groups, with the goal of maximizing the transient stability margin of the commutation bus voltage and reducing the system network loss; Where P loss is the power grid loss; λ1 and λ2 are the optimization target weight coefficients, whose values are assigned according to the optimization target bias and the order of magnitude of the optimization target value change; Q scN N is the rated capacity of a single phase regulator; min The minimum input group of AC filter; Q cN is the rated capacity of a single filter group; V max 、V min They are the upper and lower limits of the commutation bus voltage respectively.
2. The method for optimizing steady-state reactive power coordination strategy of converter stations based on data drive according to claim 1, characterized in that: Select the minimum value of the transient stability margin index of the commutation bus voltage under multiple expected severe faults (f=1,…,n) of the DC-fed grid As the transient stability margin of the commutation bus voltage, a quantitative evaluation index of the voltage transient stability margin based on multi-binary table and weighted integration is proposed The definition is as follows: Where: V N is the bus reference voltage, V(t) is the bus transient voltage under fault disturbance; t p and t′ p are the voltages below and above the pth voltage critical reference value V during the drop and recovery process of V(t), respectively. cr,p The moment of K p is(t p ,t p+1 )∪(t′ p+1 ,t′ p ) interval integral weight coefficient; when V(t) is less than V cr,1 ,…,V cr,p ,…,V cr,q When the voltage drop and recovery process is located at (t1, t2)∪(t′2, t′1),…,(t p ,t p+1 )∪(t′ p+1 ,t′ p ),…,(t q ,t′ q ) The integral weight coefficients of the time interval are K1,…,K p ,…,K q ; The weight coefficient of each interval can be based on the voltage transient response, in order to keep the voltage critical transient stability, that is, η V =0, determined by solving formula (1) step by step; Where, T cr,q For voltages lower than V cr,q The maximum acceptable duration.
3. The method for optimizing steady-state reactive power coordination strategy of converter stations based on data drive according to claim 1 is characterized in that: The process of establishing a support vector regression model for predicting the transient stability margin of the commutation bus voltage based on the multi-dimensional input feature vector by using the transient stability calculation of the power grid for anticipated severe faults includes: S1. Adjust the number of phase regulators and AC filters in the converter station and calculate the power flow. Meanwhile, perform transient stability calculations on the converged power flows after reactive power adjustment according to the expected severe fault set. S2. construct a multi-dimensional input feature vector according to the calculation results of each convergent power flow, and determine the transient stability margin of the commutation bus voltage based on the transient stability calculation results of the convergent power flow under each expected severe fault, so as to construct machine learning sample data for support vector regression modeling for predicting the transient stability margin of the commutation bus voltage based on the multi-dimensional input feature vector; S3, using the Z-score standardization method shown in formula (ii) to normalize the components of the multidimensional input feature vector and the transient stability margin of the commutation bus voltage in the data sample set, and after data preprocessing, using the classic 7:3 ratio to divide the samples into a training set and a test set; In the formula, x m,i and are the i-th sample X i The mth component and its normalized value, sample X i The vector is composed of the reactive power output of the phase regulator, the number of AC filter groups, the reactive exchange between the converter station and the DC feed-in grid, the steady-state voltage of the commutation bus, and the transient stability margin of the commutation bus voltage. and are the mean and standard deviation of the mth component respectively; S4, the training set sample X i Substitute the vector into the mathematical optimization model of the commutation bus voltage transient stability margin prediction model based on support vector machine regression shown in formula (III), use k-fold cross validation to perform grid global parameter optimization on the model parameters c and g, and finally establish the commutation bus voltage transient stability margin prediction model based on the training set data, and determine the final prediction model after evaluating the learning effect of the model based on the test set data; In the formula, α i and represents a Lagrangian operator greater than or equal to zero, X i and X j are the i-th and j-th samples of the training set, l is the number of samples in the training set, ε is the pipeline threshold that describes the insensitive deviation range between the model predicted output and the actual value of the transient stability margin of the commutation bus voltage, c is the regularization constant, and the larger the c, the greater the penalty for the sample points that exceed the ε pipeline. K(X i ,X j ) is a Gaussian radial kernel function, which is used to calculate X in the original sample space. i and X j The inner product in the high-dimensional feature space enables regression learning to be performed implicitly in the feature space, and g is the kernel parameter.
4. The method for optimizing steady-state reactive power coordination strategy of converter stations based on data drive according to claim 1, characterized in that: The solution process of the reactive power coordination strategy optimization model includes: S1. Initialize the particle swarm parameters, including setting the weight parameters c1 and c2 that reflect the particle's own experience and social experience, setting the particle's inertia weight parameter ω∈[0,1] to maintain the existing optimization speed, and setting the dimension of the optimization target search space equal to the number of reactive power optimization variables of the converter station. Set the optimization space S according to the number of AC filter groups and the rated capacity of the phase shifter. n ;vector is the optimal particle i, where the component represents the reactive output of the phase regulator of the optimal particle i, and the component Represents the number of AC filter groups of the optimal particle i, defines the number of particles as r, and in the optimal space S n The initial components of each particle are randomly generated in At the same time, the initial optimization speed of each particle is randomly generated The optimal fitness value of each particle to the objective function The optimal fitness value of the particle swarm is Fit p =∞; the maximum number of iterations of the optimal search is k max , and set the initial value of the iteration number k to 1; S2. Update the output reactive power of the condenser and AC filter of the converter station based on each optimal particle and calculate the power flow. For each converged power flow, construct a multi-dimensional input feature vector based on the power flow result and predict the temporary stability margin of the commutation bus voltage. At the same time, calculate the objective function of the optimization model shown in formula (IV) and analyze the fitness value of each particle in the current kth iteration. And judge it, if but And make Otherwise, keep the optimal fitness value of each particle With the best position unchanged, and at the same time, the optimal fitness value of each particle is maintained for each non-convergent trend With the best position unchanged, after all particles have been optimally updated, if Fit p Updated to And update The corresponding particle is the global optimal particle Otherwise keep Fit p and constant; S3, use the particle velocity and displacement update formula to calculate the particle's spatial position at the k+1th iteration. If k+1 <k max , then k=k+1, return to S2 to perform the next iterative particle optimization calculation until the number of iterations reaches k max , output the optimal solution.
5. A data-driven converter station steady-state reactive power coordination strategy optimization device, using the method described in any one of claims 1 to 4, characterized in that: include: Module 1: Based on the reactive output of the condenser phase in the converter station, the number of AC filter groups, the reactive exchange between the converter station and the DC feed-in grid, and the steady-state voltage of the converter bus, a multi-dimensional input feature vector of the regression model for predicting the transient stability margin of the converter bus voltage is constructed, and the support vector regression method is used to establish the prediction model of the transient stability margin of the converter bus voltage; The second module: Taking the maximization of the transient stability margin of the commutation bus voltage and the reduction of the system network loss as the optimization goals, the coordinated optimization objective function of the reactive output of the phase regulator and AC filter of the converter station is established to take into account both the improvement of the transient stability of the commutation bus voltage and the economic efficiency of the system operation, and additional constraints such as the number of AC filter groups in the converter station, the reactive exchange amount of the commutation bus, the reactive output of the phase regulator, and the node voltage are added to establish an optimization model for the steady-state reactive output of the converter station; The third module: Solve the steady-state reactive power optimization model of the converter station based on power flow calculation and particle swarm algorithm, specifically including: taking the output of the condenser and the number of filter groups as the optimal particles, initializing the particle swarm parameters and performing power flow analysis and calculation of the DC feed-into the power grid. For converged power flow and the power flow calculation results satisfying the additional constraints of the optimization model, a prediction model is constructed based on the power flow calculation results, mostly as input feature quantities, and the commutation bus voltage temporary stability margin prediction model and the objective function are used to update the individual particle optimal and the population optimal. Otherwise, skip directly updating the particle speed and position, and repeat this cycle until the number of iterations reaches the preset number and the optimal solution is output; Considering that the DC transmission power and the reactive voltage regulation equipment fed into the grid remain unchanged in the initial state during the steady-state operation of the power grid, and considering the constraints such as the reactive exchange of the converter station, the minimum number of filter groups of the AC filter and the reactive capacity of the phase regulator, the optimization model shown in formula (IV) is established by optimizing the steady-state reactive output of the phase regulator and the number of AC filter groups, with the goal of maximizing the transient stability margin of the commutation bus voltage and reducing the system network loss; Where P loss is the power grid loss; λ1 and λ2 are the optimization target weight coefficients, whose values are assigned according to the optimization target bias and the order of magnitude of the optimization target value change; Q scN N is the rated capacity of a single phase regulator; min The minimum input group of AC filter; Q cN is the rated capacity of a single filter group; V max 、V min They are the upper and lower limits of the commutation bus voltage respectively.
