Optimization method and device for reactive power compensation control of continuously adjustable new energy compensation filter

Abstract

The application discloses a reactive power compensation control optimization method and equipment of a continuously adjustable new energy compensation filter, relates to the technical field of reactive power compensation, and comprises the following steps: obtaining a reactive power target by analysis; taking the two-norm square of a reactive power compensation distribution weight as a weight dispersion, obtaining a first weight dispersion; generating a corresponding reactive power compensation distribution weight vector; performing adaptive reactive power compensation based on the reactive power compensation distribution weight vector, obtaining reactive power compensation data, and generating a weight dispersion evaluation model based on a neural network algorithm training, and outputting a first evaluation value; constructing a weight-compensation effect feature map according to the first evaluation value and the corresponding weight dispersion, performing image analysis, screening out a third weight dispersion, and applying the third weight dispersion to reactive power compensation, so that the problems of low reactive power distribution precision, slow response and unstable system voltage regulation of the new energy compensation filter are solved.

Description

Technical Field

[0001] This invention relates to the field of reactive power compensation technology, and more specifically, to a method and device for reactive power compensation control optimization of a continuously adjustable renewable energy compensation filter. Background Technology

[0002] A continuously adjustable renewable energy compensation filter is a device capable of adjusting the output reactive power in real time to maintain grid voltage stability. Its reactive power compensation control typically relies on optimizing the allocation weights of each compensation unit. Existing technologies mostly allocate weights based on preset weights or the capacity ratio of each unit, combined with real-time voltage monitoring to adjust the units and achieve voltage fluctuation suppression and reactive power balance. While these methods theoretically provide continuously adjustable reactive power output, they still suffer from insufficient adjustment accuracy and dynamic response in practical operation.

[0003] In existing continuously adjustable renewable energy compensation filters, both excessively high and excessively low reactive power compensation allocation weights have their drawbacks. Excessively high weights may cause some compensation units to operate under high loads for extended periods, leading to equipment overload, shortened lifespan, and increased local voltage fluctuations. Conversely, excessively low weights may result in insufficient compensation contribution, slow system response, and difficulty in timely adjusting voltage deviations, thereby affecting reactive power regulation and grid stability.

[0004] The drawbacks of excessively high weighting are that the reactive power output of each compensation unit approaches or reaches its maximum available capacity, increasing the operational pressure on the units and easily triggering capacity limit protection. Simultaneously, high weighting may cause overvoltage phenomena at local nodes, with large deviations and long durations, increasing the risk of local grid fluctuations. Although the total reactive power error may decrease rapidly in the short term, the large fluctuation amplitude reduces the dynamic performance of the filter and the stability of the system.

[0005] The disadvantages of excessively low weight allocation are that each compensation unit has insufficient output power, low utilization rate, and limited compensation contribution, resulting in a slower overall system response and a longer voltage deviation convergence time. Long-term low weight allocation may cause the total reactive power error to persist, failing to reach the allowable range within the specified time, thereby reducing the filter's compensation accuracy and the stability of the grid voltage, while also affecting the load balancing effect.

[0006] Existing technologies cannot effectively determine the optimal reactive power compensation allocation weights, resulting in insufficient reactive power compensation accuracy and unsatisfactory dynamic response in the actual operation of continuously adjustable renewable energy compensation filters, and increasing the uncertainty and safety risks of system operation.

[0007] To address the above problems, this invention proposes a solution. Summary of the Invention

[0008] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a reactive power compensation control optimization method and device for a continuously adjustable renewable energy compensation filter. By using continuously adjustable optimization based on the capacity and weight dispersion of the compensation unit and neural network evaluation, the problem of poor control effect caused by uneven weight distribution in traditional methods can be solved.

[0009] To achieve the above objectives, the present invention provides the following technical solution:

[0010] A reactive power compensation control optimization method for a continuously adjustable renewable energy compensation filter includes the following steps: real-time acquisition of reactive power, and analysis to obtain the reactive power target based on preset voltage control requirements; setting the squared L2 norm of the reactive power compensation allocation weights as the weight dispersion, and obtaining the first weight dispersion; generating several second weight dispersions based on the first weight dispersion using random numbers, and generating corresponding reactive power compensation allocation weight vectors; performing adaptive reactive power compensation based on the reactive power compensation allocation weight vectors, obtaining reactive power compensation data, and training a weight dispersion evaluation model based on a neural network algorithm, outputting the first evaluation value; constructing a weight-compensation effect feature map based on the first evaluation value and the corresponding weight dispersion, and performing image analysis to filter out the third weight dispersion, which is then applied to reactive power compensation; the generation of the corresponding reactive power compensation allocation weight vectors is based on the binary allocation construction to limit and solve each value of the reactive power compensation allocation weight vectors.

[0011] In a preferred embodiment, the real-time acquisition of reactive power, combined with preset voltage control requirements, and the analysis to obtain the reactive power target, specifically involves: using the branch reactive power at the grid connection point and branch nodes, where the preset voltage control requirements include the allowable voltage fluctuation range and the voltage target; performing time synchronization, noise filtering, and anomaly removal on the acquired reactive power data to obtain the real-time net reactive power; calculating the current voltage deviation based on the preset voltage control requirements and the real-time net reactive power, and deriving the reactive power adjustment amount required to meet the voltage target by combining the system reactive power-voltage sensitivity coefficient; and superimposing the real-time net reactive power and the reactive power adjustment amount to obtain the target reactive power compensation amount as the reactive power target.

[0012] In a preferred embodiment, the step of setting the L2 norm squared of the reactive power compensation allocation weight as the weight dispersion and obtaining the first weight dispersion specifically involves: obtaining the available compensation capacity of each compensation unit and calculating the first capacity ratio; obtaining the first weight vector based on the normalized comprehensive score according to the first capacity ratio; and calculating the L2 norm squared of the first weight vector as the first weight dispersion.

[0013] In a preferred embodiment, the step of generating several second weight dispersions based on random numbers based on the first weight dispersion specifically involves: generating several computer random numbers based on the first weight dispersion; and superimposing the computer random numbers onto the first weight dispersion to obtain several second weight dispersions.

[0014] In a preferred embodiment, the generation of the corresponding reactive power compensation allocation weight vector is based on the limitation and solution of each value of the reactive power compensation allocation weight vector by constructing a binary allocation. Specifically, the compensation units are sorted according to the available capacity of the branches; the number of potential concentrated groups is obtained and a binary allocation pattern is generated; the second weight value corresponding to the binary value is calculated based on the weight dispersion using a closed formula, and a second weight vector is generated based on the sorting of compensation units; the number of concentrated groups is adjusted based on the upper bound constraint of the capacity of each unit, thereby correcting the second weight vector and generating a third weight vector.

[0015] In a preferred embodiment, the adaptive reactive power compensation based on the reactive power compensation weight vector, acquiring reactive power compensation data, and training a weight dispersion evaluation model based on a neural network algorithm to output a first evaluation value, specifically involves: the reactive power compensation data including the proportion of reactive power output of each compensation unit to its available capacity, the excess amplitude and duration of local node voltage deviation, and the amplitude and convergence time of the total reactive power error; several second weight dispersions corresponding to several sets of reactive power compensation data; using the reactive power compensation data as input features of the neural network model to train the neural network to learn the influence law of weight dispersion on the compensation effect; and using the trained neural network model to predict the evaluation value corresponding to each set of weight vectors as the first evaluation value.

[0016] In a preferred embodiment, the step of constructing a weight-compensation effect feature map based on the first evaluation value and the corresponding weight dispersion, and performing image analysis to filter out the third weight dispersion for application to reactive power compensation, specifically involves: combining each third weight dispersion with the corresponding first evaluation value obtained after reactive power compensation to construct a two-dimensional relationship vector; mapping the two-dimensional relationship vectors of several third weight dispersions onto a two-dimensional plane and performing difference smoothing to obtain the weight-compensation effect feature map; selecting a filtering region in the weight-compensation effect feature map according to the pre-obtained allocation requirements, and selecting the third weight dispersion corresponding to the point with the highest first evaluation value in the filtering region as the control target; generating a reactive power compensation allocation weight vector based on the control target and applying it to reactive power compensation.

[0017] A reactive power compensation control optimization system for a continuously adjustable renewable energy compensation filter includes a data acquisition module, an initial weight dispersion acquisition module, a filter set generation module, a monitoring and evaluation module, a control target selection module, and a conversion module. The data acquisition module collects reactive power in real time and analyzes and obtains the reactive power target based on preset voltage control requirements. The initial weight dispersion acquisition module uses the squared L2 norm of the reactive power compensation allocation weights as the weight dispersion to obtain a first weight dispersion. The filter set generation module generates several second weight dispersions based on the first weight dispersion using random numbers and generates corresponding reactive power compensation allocation weight vectors. The monitoring and evaluation module performs adaptive reactive power compensation based on the reactive power compensation allocation weight vectors, acquires reactive power compensation data, trains a weight dispersion evaluation model based on a neural network algorithm, and outputs a first evaluation value. The control target selection module constructs a weight-compensation effect feature map based on the first evaluation value and the corresponding weight dispersion, performs image analysis to select a third weight dispersion for application in reactive power compensation. The conversion module is used to limit and solve the values ​​of the reactive power compensation allocation weight vector based on a binary allocation construction.

[0018] An electronic device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores a computer program executable by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to execute a reactive power compensation control optimization method for a continuously adjustable renewable energy compensation filter.

[0019] A computer-readable storage medium storing a computer program, which, when executed by a processor, implements a reactive power compensation control optimization method for a continuously adjustable renewable energy compensation filter.

[0020] The technical effects and advantages of the continuously adjustable renewable energy compensation filter reactive power compensation control optimization method and equipment of this invention are as follows:

[0021] 1. This invention, by real-time acquisition of reactive power and key node voltage data from each compensation unit, and combining this with preset voltage control requirements and the system's reactive power-voltage sensitivity coefficient, derives the reactive power adjustment amount required to meet the voltage target, thus forming a real-time reactive power target. Based on this, the capacity ratio of each compensation unit is calculated, and an initial weight vector is generated through normalized comprehensive scoring. The weight dispersion is further calculated to provide a continuously adjustable reference for candidate weight schemes. The innovation of this technology lies in combining real-time measurement data with capacity constraints, sensitivity coefficients, and dispersion indices, enabling the reactive power target calculation to reflect the system's dynamic state and provide a continuously adjustable and quantifiable weight exploration basis for subsequent optimization, thereby improving the flexibility and accuracy of reactive power allocation schemes.

[0022] 2. This invention generates a second set of weighted discreteness based on a first weighted discreteness, which consists of multiple random disturbances. A corresponding reactive power compensation weight vector is then constructed using binary allocation. This is corrected by considering the upper bound constraint of the compensation unit capacity, forming a third set of weighted vectors. Through adaptive compensation, features such as the output occupancy rate of each unit, the amplitude and duration of local voltage deviation, the amplitude of total reactive power error, and the convergence time are collected. A neural network model is trained to generate a weight-compensation effect mapping, and the optimal third weighted discreteness is selected to generate the final control weights. The innovation of this technology lies in using continuously adjustable discrete disturbances and closed-loop binary allocation combined with neural network evaluation to achieve feature-quantized weight optimization. This allows for optimal reactive power compensation while ensuring unit capacity safety, improving overall system stability, dynamic response performance, and adjustment accuracy. Attached Figure Description

[0023] Figure 1 This is a flowchart illustrating the reactive power compensation control optimization method for the continuously adjustable renewable energy compensation filter of the present invention.

[0024] Figure 2 This is a schematic diagram of the reactive power compensation control optimization system of the continuously adjustable renewable energy compensation filter of the present invention. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0026] Example 1, Figure 1 The present invention provides a reactive power compensation control optimization method for a continuously adjustable renewable energy compensation filter, comprising the following steps:

[0027] S1 collects reactive power and status variables in real time, and analyzes and obtains reactive power targets in combination with preset voltage control requirements.

[0028] In this embodiment, the real-time acquisition of reactive power, combined with preset voltage control requirements, and the analysis to obtain the reactive power target, specifically involves:

[0029] Based on the branch reactive power of the grid connection point and branch nodes, the preset voltage control requirements include the allowable voltage fluctuation range and the voltage target.

[0030] The collected reactive power data is time-synchronized, noise-filtered, and anomaly-removed to obtain real-time net reactive power.

[0031] The current voltage deviation is calculated based on the preset voltage control requirements and real-time net reactive power, and the reactive power regulation required to meet the voltage target is derived by combining the system reactive power-voltage sensitivity coefficient.

[0032] The real-time net reactive power is superimposed with the reactive power regulation to obtain the target reactive power compensation amount as the reactive power target.

[0033] It should be noted that the minimum set of measurements used to generate the reactive power target in real time only includes the grid connection point bus voltage, the system net reactive power, and the available compensation capacity of each compensation unit, in order to ensure that the subsequent reactive power adjustment calculation has reliable input and meets the capacity constraints.

[0034] It should be noted that the preset voltage control requirements should clearly define the voltage target and allowable fluctuation range. The voltage target can be issued by the superior dispatch system, specified by the grid connection procedure, or set by the field operation strategy to ensure that there is a unified reference for reactive power target calculation.

[0035] It should be noted that the system reactive power-voltage sensitivity coefficient can be obtained through offline power flow linearization, online small disturbance test, or recursive least squares identification based on time-series measurement, and is used to establish a linear approximate mapping relationship between voltage deviation and reactive power regulation.

[0036] It should be noted that the collected reactive power data needs to undergo time synchronization, noise filtering, and anomaly removal to eliminate the interference of measurement errors and transient impacts on the compensation calculation. Otherwise, the compensation amount will fluctuate significantly in a short period of time, resulting in frequent filter adjustments and shortened lifespan.

[0037] It should be noted that the final reactive power target is the result of superimposing the net reactive power with the required reactive power adjustment. This target directly serves as the input for subsequent compensation capacity allocation and control command generation, and plays a decisive role in the reactive power compensation accuracy of the continuously adjustable renewable energy compensation filter.

[0038] S2, let the square of the second norm of the reactive power compensation allocation weight be used as the weight dispersion, and obtain the first weight dispersion.

[0039] In this embodiment, the first weight dispersion is obtained by taking the squared L2 norm of the reactive power compensation allocation weight as the weight dispersion, specifically as follows:

[0040] Obtain the available compensation capacity of each compensation unit and calculate the first capacity ratio;

[0041] The first weight vector is obtained based on the first capacity ratio and the normalized comprehensive score;

[0042] The squared L2 norm of the first weight vector is used as the first weight dispersion.

[0043] It should be noted that the following is a feasible example of calculating the first capacity ratio:

[0044] ;

[0045] In the formula, The capacity ratio of the i-th compensation unit. Let i be the maximum reactive power capacity that the i-th compensation unit can allocate in the current state. Let j be the maximum reactive power capacity that the j-th compensation unit can allocate in the current state. This represents the total number of compensation units.

[0046] It should be noted that the following is a feasible example of calculating the first weight vector:

[0047] ;

[0048] In the formula, Let be the first weight of the i-th compensation unit, and construct a first weight vector by sorting several first weights according to their serial numbers.

[0049] It should be noted that the following is a feasible example of calculating the first weight dispersion:

[0050] ;

[0051] In the formula, The first weighted dispersion.

[0052] It should be noted that the available compensation capacity refers to the reactive power compensation capacity that each compensation unit can actually output under the current operating state and equipment constraints. It is usually calculated by monitoring the current operating position, switch status and rated capacity of the compensator to ensure that the calculation results can reflect the real-time adjustable compensation capability.

[0053] It should be noted that the first capacity ratio refers to the ratio of the available compensation capacity of each compensation unit to the total available compensation capacity of all compensation units. This ratio can characterize the contribution ratio of each unit to the overall compensation capacity, providing a basic quantitative indicator for subsequent weight calculation.

[0054] It should be noted that the normalized comprehensive score is calculated by standardizing the capacity ratio of each compensation unit and combining multiple factors such as adjustment response speed and position voltage sensitivity to ensure that the obtained weight vector reflects both the capacity ratio and the compensation effect and dynamic response performance.

[0055] It should be noted that the L2 norm squared of the weight vector, as the weight dispersion, can measure the degree of concentration or dispersion of the capacity allocation of each compensation unit. The larger the value, the more concentrated the allocation, and the smaller the value, the more uniform the allocation. This index can provide a continuously adjustable reference benchmark for the subsequent generation of candidate allocation schemes.

[0056] S3, based on the first weight dispersion, generate several second weight dispersions based on random numbers, and generate the corresponding reactive power compensation allocation weight vector;

[0057] In this embodiment, the step of generating several second weight dispersions based on random numbers based on the first weight dispersion specifically involves:

[0058] Based on the first weighted dispersion, generate several computer-generated random numbers;

[0059] Computer-generated random numbers are superimposed on the first weighted dispersion to obtain several second weighted dispersions.

[0060] In this embodiment, the generation of the corresponding reactive power compensation allocation weight vector is based on the binary allocation construction, which limits and solves for each value of the reactive power compensation allocation weight vector, specifically as follows:

[0061] The compensation units are sorted according to the available capacity of the branch lines;

[0062] Obtain the number of potential clusters and generate a binary assignment pattern;

[0063] The second weight value corresponding to the binary value is calculated based on the weight dispersion using a closed formula. And based on the sorting of the compensation units, a second weight vector is generated;

[0064] The number of centralized groups is adjusted based on the upper bound constraint of the capacity of each unit, thereby correcting the second weight vector and generating the third weight vector.

[0065] It should be noted that the aforementioned binary allocation mode is specifically as follows:

[0066] Let m be the number of potential centralized groups in N compensation units. Then, set the first m units with the largest capacity as centralized groups and the Nm units as decentralized groups to obtain a binary allocation mode.

[0067] It should be noted that, for a given potential set of groups, the following example illustrates how to calculate the binary weight allocation using a closed-form formula so that the total weight sum and the weight dispersion simultaneously satisfy the first constraint:

[0068] ;

[0069] ;

[0070] In the formula, For the number of centralized groups, Assign weights to the corresponding groups. The total number of compensation units, Assign weights to the distributed groups. This represents the weighted dispersion.

[0071] It should be noted that the feasible calculation example for the second weight value corresponding to the binary value using the closed-form formula is as follows:

[0072] ;

[0073] ;

[0074] Where a is greater than or equal to b, and has a real root only when S is greater than 1 / N, the closed formula combined with the first constraint is used to solve for the weight allocation.

[0075] It should be noted that the adjustment of the number of ensemble groups based on the upper bound constraint of each unit's capacity, thereby correcting the second weight vector and generating the third weight vector, is as follows:

[0076] The following is a calculation example for verifying whether the upper bound of the capacity of each cell satisfies the constraints:

[0077] ;

[0078] ;

[0079] In the formula, and Both are upper bound constraints on cell capacity; i belongs to the concentrated group, and j belongs to the dispersed group. This represents the total reactive power that needs to be adjusted for this reactive power compensation.

[0080] It should be noted that if there are any exceptions, the binary weight allocation will be recalculated after adjusting m.

[0081] It should be noted that the following is an example of a feasible arithmetic expression for the third weight vector:

[0082] .

[0083] It should be noted that the first weight dispersion refers to the squared L2 norm calculated based on the initial weight vector. It reflects the degree of concentration or dispersion of the weights of each compensation unit and serves as the benchmark value for generating subsequent candidate dispersions. This index allows for continuous and adjustable weight exploration while maintaining reasonable initial values.

[0084] It should be noted that computer-generated random numbers are pseudo-random sequences generated by algorithms. These sequences are used to introduce perturbations into the first weighted dispersion, forming several second weighted dispersions. The random numbers can be generated using a uniform or Gaussian distribution, and upper and lower limits can be set to ensure that the generated second weighted dispersions fall within the feasible range for engineering applications.

[0085] It should be noted that the second weight dispersion refers to the candidate weight dispersion set after adding random perturbation on the basis of the first weight dispersion. It reflects a variety of possible weight allocation modes and is used to generate reactive power compensation allocation weight vectors and perform effect screening.

[0086] It should be noted that the available capacity of a branch refers to the actual reactive power compensation capacity that each compensation unit can provide under the current operating conditions. It is usually calculated by the rated capacity, temperature limit, and real-time operating level of the monitoring device, and is used to sort the compensation units so that units with larger capacities can take on a larger weight during centralized allocation.

[0087] It should be noted that the number of potential concentrated groups refers to the number of compensation units selected as "concentrated groups" and given higher weights under the binary allocation mode. Choosing an appropriate number of concentrated groups can adjust the dispersion of the weight vector, thereby controlling the concentration or uniformity of the allocation and affecting the final reactive power compensation effect.

[0088] It should be noted that the binary allocation mode is a method for constructing weight vectors, which divides the compensation units into centralized and decentralized groups, assigning two different weight values ​​to each. This method can quickly generate weight vectors that meet specified dispersion requirements while maintaining the controllability and continuous adjustability of the allocation.

[0089] It should be noted that the second weight vector is a complete weight vector generated based on the calculated values ​​of a and b, combined with the sorting of compensation units. By assigning a to the centralized group of units with larger capacity and b to the decentralized group of units, the weight vector is ensured to satisfy both the dispersion target and the unit capacity constraints and the actual operating capability of the system.

[0090] It should be noted that the capacity upper bound constraint means that the weight of each compensation unit multiplied by the total reactive power cannot exceed its available compensation capacity. If the generated second weight vector violates this constraint, the number of centralized groups needs to be adjusted or the weight vector needs to be corrected to ensure that all units will not be overloaded during actual allocation, thereby ensuring the safety and stability of reactive power compensation of the continuously adjustable renewable energy compensation filter.

[0091] It should be noted that the third weight vector is the final weight vector obtained after correcting the second weight vector for capacity upper bound constraints. This third weight vector is used as a candidate scheme input into subsequent reactive power compensation control steps to calculate the actual reactive power allocation of each unit. This vector directly affects the accuracy, response speed, and system stability of reactive power compensation.

[0092] S4. Adaptive reactive power compensation is performed based on the reactive power compensation weight vector allocation to obtain reactive power compensation data, and a weight dispersion evaluation model is generated based on the neural network algorithm to output the first evaluation value.

[0093] In this embodiment, the adaptive reactive power compensation based on the reactive power compensation allocation weight vector, obtaining reactive power compensation data, and training a weight dispersion evaluation model based on a neural network algorithm to output a first evaluation value, specifically involves:

[0094] The reactive power compensation data includes the proportion of reactive power output of each compensation unit to its available capacity, the excess amplitude and duration of local node voltage deviation, and the amplitude and convergence time of the total reactive power error.

[0095] Several second-weighted dispersions correspond to several sets of reactive power compensation data;

[0096] The reactive power compensation data is used as the input feature of the neural network model to train the neural network to learn the influence of weight dispersion on the compensation effect.

[0097] The trained neural network model is used to predict the evaluation value corresponding to each set of weight vectors, which is then used as the first evaluation value.

[0098] It should be noted that the method for obtaining the proportion of reactive power output of each compensation unit to its available capacity is by real-time monitoring of the reactive power output of each compensation unit, and then comparing it with the maximum reactive power capacity allowed to be output by that unit under the current operating conditions, thereby obtaining the utilization rate index. This technical point involves high-precision measurement of power sensors, data acquisition and filtering processing, and considers the correction of the maximum capacity by factors such as temperature, equipment aging or safety limitations, so as to ensure the accuracy and operability of the proportion value.

[0099] It should be noted that the method for obtaining the magnitude and duration of local node voltage deviations involves deploying voltage sensors at key nodes of the power grid, continuously collecting voltage values, comparing them with the node's set allowable voltage range, and recording the deviation range and duration. This technique involves time synchronization of voltage sampling, filtering and noise reduction, outlier removal, and an event recognition algorithm for deviations exceeding thresholds to ensure accurate quantification of voltage deviation characteristics.

[0100] It should be noted that the magnitude and convergence time of the total reactive power error are obtained by summing the output reactive power of all compensation units and comparing it with the target reactive power compensation amount, calculating the absolute value of the total error, and recording the time required for the error to decay from the initial value to the allowable error range. This technique involves dynamic acquisition of the compensation amount, error calculation algorithm, signal filtering, and response time analysis to ensure that the error characteristics can truly reflect the overall compensation effect of the system.

[0101] It should be noted that the ratio of reactive power output to available capacity for each compensation unit is obtained by real-time monitoring of the actual reactive power output of each compensation unit and calculating the ratio in conjunction with its maximum available capacity under design or operational limitations. If the allocation weight is too high, this ratio may approach or reach its upper limit, reflecting potential unit overload and high operational pressure; conversely, if the allocation weight is too low, the ratio is small, indicating limited unit compensation contribution and insufficient response. This characteristic not only reflects the unit load balance but also directly affects the overall reactive power compensation effect of the continuously adjustable renewable energy compensation filter.

[0102] It should be noted that the method for obtaining the magnitude and duration of local node voltage deviations involves deploying voltage measuring devices at key nodes of the power grid to continuously collect node voltage data and compare it with the allowable voltage fluctuation range, recording the deviation magnitude and its duration. If the weight allocation is too high, some nodes may experience overvoltage with large magnitude and long duration, affecting system stability; if the weight allocation is too low, the voltage deviation may converge slowly and fail to reach the target quickly, leading to a delay in reactive power compensation response. This feature can be used to evaluate the impact of weight allocation on the local voltage regulation effect.

[0103] It should be noted that the magnitude and convergence time of the total reactive power error are obtained by summarizing the reactive power output of all compensation units, calculating the difference between this value and the target compensation amount, and monitoring the time required for the error to decrease from its initial value to the allowable range. If the weight is too high, some units may be overloaded, causing the error to decrease in the short term but with large fluctuations, potentially leading to control instability. If the weight is too low, the total error converges slowly, reducing the overall system response capability and affecting the compensation accuracy and dynamic performance of the continuously adjustable renewable energy compensation filter. This characteristic reflects the overall compensation coordination of the system.

[0104] It should be noted that several second weight dispersions correspond to several sets of reactive power compensation data, meaning that each candidate weight allocation scheme generates a set of feature data through adaptive compensation for subsequent performance evaluation. This mapping quantifies the feature changes caused by high and low weights, facilitating the neural network model's learning of the compensation effect patterns under different dispersions, thus providing a basis for optimal weight selection.

[0105] It should be noted that when using reactive power compensation data as input features to a neural network model, it is necessary to integrate the capacity utilization rate of each unit, the amplitude and duration of local node voltage deviation, and the amplitude and convergence time of the total reactive power error into a feature vector, which is then input into the neural network for training to learn the comprehensive impact of weight dispersion on the compensation effect. Existing mature neural network training and feature standardization techniques can be used directly and do not require further explanation.

[0106] It should be noted that when using the trained neural network model to predict the first evaluation value, the compensation features corresponding to each set of weight vectors are input into the model, and a numerical index is output to quantify the allocation effect. This evaluation value can reflect the combined impact of high and low weights on unit load, voltage regulation, and total reactive power error of the system, providing a reference for selecting the optimal weight vector and optimizing the reactive power compensation of the continuously adjustable renewable energy compensation filter.

[0107] S5. Construct a weight-compensation effect feature map based on the first evaluation value and the corresponding weight dispersion, and perform image analysis to select the third weight dispersion, which is then applied to reactive power compensation.

[0108] In this embodiment, the step of constructing a weight-compensation effect feature map based on the first evaluation value and the corresponding weight dispersion, and then performing image analysis to filter out the third weight dispersion for application in reactive power compensation, specifically involves:

[0109] Each third weight dispersion is combined with the first evaluation value obtained after the corresponding application reactive power compensation to construct a two-dimensional relationship vector;

[0110] Several two-dimensional relation vectors with third weight dispersion are mapped to a two-dimensional plane and then subjected to difference smoothing to obtain a weight-compensation effect feature map.

[0111] Based on the pre-obtained allocation requirements, a filtering region is selected in the weight-compensation effect feature map, and the third weight dispersion corresponding to the point with the highest first evaluation value in the filtering region is selected as the control target.

[0112] A reactive power compensation allocation weight vector is generated based on the control objective and applied to reactive power compensation.

[0113] It should be noted that each third weight dispersion and the corresponding first evaluation value obtained after reactive power compensation are combined to form a two-dimensional relationship vector. This means that the dispersion of the weight allocation is used as the horizontal axis, and the evaluation value obtained after applying the dispersion for reactive power compensation is used as the vertical axis, thereby quantifying the relationship between the weight dispersion and the compensation effect for subsequent analysis and screening.

[0114] It should be noted that mapping several two-dimensional relation vectors of third weight dispersion onto a two-dimensional plane and performing difference smoothing means forming a distribution map of discrete data points on a two-dimensional plane, and smoothing the values ​​between discrete points through interpolation algorithms (such as bilinear interpolation or spline interpolation) to obtain a continuous weight-compensation effect feature map, which facilitates observation of the trend of compensation effect with dispersion and finding the optimal point.

[0115] It should be noted that selecting the third weight dispersion corresponding to the point with the highest first evaluation value in the screening area as the control target means finding the weight dispersion that can provide the best compensation effect within the candidate area that meets the allocation requirements, thereby guiding the generation of the final reactive power compensation allocation weight vector and ensuring the reactive power compensation accuracy and dynamic performance of the continuously adjustable new energy compensation filter.

[0116] It should be noted that generating a reactive power compensation allocation weight vector based on the control objective and applying it to reactive power compensation means that, based on the selected optimal weight dispersion, the final weight value of each compensation unit is generated through binary allocation or closed-form calculation formula, and this weight vector is applied to the actual reactive power compensation control, so that the reactive power output of each compensation unit works according to the optimal allocation ratio, thereby optimizing the system voltage stability and overall reactive power regulation performance.

[0117] It should be noted that the pre-acquired allocation requirements refer to the reactive power compensation targets and constraints determined during the system design or operation planning phase. These include the system's allowable range for voltage fluctuations, capacity limitations of each compensation unit, system response speed requirements, and total reactive power regulation targets. These requirements guide the selection of weight dispersion and the generation of the final allocation vector, ensuring that the compensation effect meets operational safety and performance indicators.

[0118] It should be noted that each third weight dispersion and the corresponding first evaluation value obtained after reactive power compensation form a two-dimensional relationship vector. This means that the candidate weight dispersion is used as the horizontal axis and the evaluation value obtained after applying the dispersion for compensation is used as the vertical axis, thereby quantitatively describing the impact of dispersion on the compensation effect and providing basic data for feature map construction.

[0119] It should be noted that mapping several two-dimensional relation vectors of third-weight discreteness to a two-dimensional plane and performing difference smoothing processing means presenting the discrete points as a distribution on the two-dimensional plane, and using interpolation algorithms such as bilinear interpolation or spline interpolation to smooth the values ​​between discrete points to generate a continuous surface, thereby obtaining a weight-compensation effect feature map, which is convenient for intuitive analysis and optimal point identification.

[0120] It should be noted that selecting the third weight dispersion corresponding to the point with the highest first evaluation value in the screening area as the control target means finding the weight dispersion that can provide the best reactive power compensation effect in the candidate area, which is used to guide the generation of the final weight vector and ensure that the continuously adjustable new energy compensation filter achieves accurate and stable reactive power regulation in actual operation.

[0121] It should be noted that generating a reactive power compensation allocation weight vector based on the control objective and applying it to reactive power compensation refers to generating the final weights of each compensation unit through binary allocation or closed-loop calculation methods using the selected optimal dispersion. This process actually adjusts the reactive power output of each unit, thereby optimizing system voltage stability, load balancing, and overall reactive power regulation accuracy. This process relies on mature technologies such as upper bound constraints on compensation unit capacity, branch sorting, and adjustment of the number of centralized groups.

[0122] Example 2, Figure 2 This invention presents a reactive power compensation control optimization system for a continuously adjustable renewable energy compensation filter, comprising a data acquisition module, an initial weight dispersion acquisition module, a filter set generation module, a monitoring and evaluation module, a control target selection module, and a conversion module. The data acquisition module collects reactive power in real time and analyzes and obtains the reactive power target based on preset voltage control requirements. The initial weight dispersion acquisition module uses the squared L2 norm of the reactive power compensation allocation weights as the weight dispersion to obtain a first weight dispersion. The filter set generation module generates several second weight dispersions based on the first weight dispersion using random numbers and generates corresponding reactive power compensation allocation weight vectors. The monitoring and evaluation module performs adaptive reactive power compensation based on the reactive power compensation allocation weight vectors, acquires reactive power compensation data, trains a weight dispersion evaluation model based on a neural network algorithm, and outputs a first evaluation value. The control target selection module constructs a weight-compensation effect feature map based on the first evaluation value and the corresponding weight dispersion, performs image analysis to select a third weight dispersion for application in reactive power compensation. The conversion module is used to limit and solve the values ​​of the reactive power compensation allocation weight vector based on a binary allocation construction.

[0123] Furthermore, it also includes an electronic device comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores a computer program executable by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to execute a reactive power compensation control optimization method for a continuously adjustable renewable energy compensation filter.

[0124] Furthermore, it also includes a computer-readable storage medium storing a computer program that, when executed by a processor, implements a reactive power compensation control optimization method for a continuously adjustable renewable energy compensation filter.

Claims

1. A reactive power compensation control optimization method for a continuously adjustable renewable energy compensation filter, characterized in that, Includes the following steps: The squared L2 norm of the reactive power compensation allocation weights is used as the first weight dispersion. Multiple second weight dispersions and corresponding reactive power compensation allocation weight vectors are randomly generated based on the first weight dispersion, and the corresponding weight vectors are solved using the binary allocation method. Compensation is performed using weight vectors, and an evaluation model is trained based on the compensated data to output an evaluation value. A feature map is constructed based on the evaluation value and the weight dispersion. The optimal third weight dispersion is selected through image analysis for actual compensation. The squared L2 norm of the reactive power compensation allocation weights is used as the first weight dispersion, specifically: Obtain the available capacity of each compensation unit and calculate its capacity ratio as the first capacity ratio; The first capacity ratio is normalized and comprehensively scored to generate an initial first weight vector. Calculate the squared L2 norm of the first weight vector and define it as the first weight dispersion to quantify the balance of weight allocation; The step of solving for the corresponding weight vector using the binary allocation method includes: They are sorted according to the available capacity of each compensation unit; Determine the number of potential clusters and generate a binary allocation pattern accordingly; Based on the weight dispersion, the weight values ​​corresponding to the binary allocation mode are calculated by closed-loop solution, and a second weight vector is generated according to the compensation unit sorting. Based on the upper bound constraints of each unit capacity, the number of the centralized groups is adjusted and the second weight vector is modified to output the third weight vector that finally satisfies all constraints.

2. The reactive power compensation control optimization method for the continuously adjustable renewable energy compensation filter according to claim 1, characterized in that, The compensation using weight vectors also includes an analysis step that considers reactive power targets, specifically: Collect reactive power data at grid connection points and branch nodes; Data processing is performed on reactive power data to obtain real-time net reactive power; Based on the preset voltage control requirements and real-time net reactive power, combined with the system's reactive power-voltage sensitivity, the required reactive power regulation is calculated. The real-time net reactive power is superimposed with the reactive power adjustment to generate the final reactive power target.

3. The reactive power compensation control optimization method for the continuously adjustable renewable energy compensation filter according to claim 2, characterized in that, Multiple second weight dispersions are randomly generated based on the first weight dispersion, specifically as follows: Based on the first weighted dispersion, generate several computer-generated random numbers; Computer-generated random numbers are superimposed on the first weighted dispersion to obtain several second weighted dispersions.

4. The reactive power compensation control optimization method for the continuously adjustable renewable energy compensation filter according to claim 3, characterized in that, The process of using weight vectors for compensation and training an evaluation model based on the compensated data to output an evaluation value is as follows: Collect multiple sets of weight dispersion and their corresponding reactive power compensation data; Using reactive power compensation data as input features, an offline neural network model is trained to learn the nonlinear mapping law from weight dispersion to compensation effect. Using the trained neural network model, predict the comprehensive evaluation value corresponding to any weight vector online.

5. The reactive power compensation control optimization method for the continuously adjustable renewable energy compensation filter according to claim 4, characterized in that, The process involves constructing a feature map based on the evaluation value and weight dispersion, and then using image analysis to select the optimal third weight dispersion for actual compensation. Specifically: The dispersion of each weight is combined with its neural network evaluation value into a two-dimensional relation vector, and then mapped onto a two-dimensional plane to generate a weight-compensation effect feature map; According to the preset allocation requirements, a screening region is defined in the feature map, and the weight dispersion corresponding to the best evaluation point in the region is selected as the final control target. A reactive power compensation allocation weight vector is generated based on the control objective and applied to the actual system control.

6. The reactive power compensation control optimization method for the continuously adjustable renewable energy compensation filter according to claim 5, characterized in that, The binary allocation method is achieved by numerically limiting and constraining the values ​​of each element in the reactive power compensation allocation weight vector.

7. The reactive power compensation control optimization method for the continuously adjustable renewable energy compensation filter according to claim 6, characterized in that, The construction of the weight-compensation effect feature map includes: forming a scatter plot with weight dispersion as the horizontal axis and the evaluation value output by the neural network as the vertical axis, and selecting the Pareto optimal weight dispersion region through cluster analysis.

8. An electronic device, characterized in that, The electronic device includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the reactive power compensation control optimization method for the continuously adjustable renewable energy compensation filter as described in any one of claims 1 to 7.

Images